This patch includes a number of improvements:

 * Material page routes now put a repaint boundary inside their transition so
   they don't repaint during the transition.
 * Heroes that are on a quest now get a repaint boundary so we repaint them
   individually.
 * I've hoisted the transparent material for the product items up in the widget
   tree, which doesn't affect performance but makes the ink splashes reach the
   edge of the product cards.
 * I've changed the repaint rainbow visualization to make it easier to see
   what's going on.

Previously, we considered them one operation in the compositor, which didn't
trigger caching. Now we have a way to explicitly hint that the compositor
should cache a layer.

Fixes #4761

We now stop updating semantics when there are no remaining clients.

Also, make SemanticsOwner into a real class and use it instead of a static in
several places.

Now that protected can be accessed from inside the same library, we can
use protected in a number of new places and we can remove some
trampolines we were using to work around its previous semantics.

The new `empty_statements` lint (in the next DEV roll), flags these empty statements.  Harmless, except the one in `basic.dart` that fixes a real-life bug. :)

Fixes #4425

Also, make sure that the parent is notified when they change.

Fixes #2298

Also, add more dartdoc to widgets.dart.

This introduces the key parts of a paginated data table, not including
the built-in pagination features.

* Provide more data for the data table demo, so there's data to page.

* Introduce a ChangeNotifier class which abstracts out
  addListener/removeListener/notifyListeners. We might be able to use
  this to simplify existing classes as well, though this patch doesn't
  do that.

* Introduce DataTableSource, a delegate for getting data for data
  tables. This will also be used by ScrollingDataTable in due course.

* Introduce PaginatedDataTable, a widget that wraps DataTable and only
  shows N rows at a time, fed by a DataTableSource.

There are many more places we can use this annotation. This patch starts out
small to test the waters.

We used to use this mechanism to paint scrollbars, but it's awkward. The
new approach to scrollbars is much cleaner. This mechanism has no other
clients, so we should remove it.

There are still many more left before rendering.dart is complete.

We now explain what coordinate system each argument is expected to be in.

Fixes #4205

For consistency.

Fixes #4142

This makes it possible to substitute 'flutter run' for 'flutter test'
and actually watch a test run on a device.

For any test that depends on flutter_test:

1. Remove any import of 'package:test/test.dart'.

2. Replace `testWidgets('...', (WidgetTester tester) {`
      with `testWidgets('...', (WidgetTester tester) async {`

3. Add an "await" in front of calls to any of the following:
    * tap()
    * tapAt()
    * fling()
    * flingFrom()
    * scroll()
    * scrollAt()
    * pump()
    * pumpWidget()

4. Replace any calls to `tester.flushMicrotasks()` with calls to
   `await tester.idle()`.

There's a guarding API that you can use, if you have particularly
complicated tests, to get better error messages. Search for
TestAsyncUtils.

It's safe to remove the unneeded `void`s from setters since the blocking issues in the
`always_declare_return_types` lint have been fixed (https://github.com/dart-lang/linter/).  We can also safely flip the bit on  `avoid_return_types_on_setters`.

* Track the worst frame in perf tests

* Use backticks in "/// Returns `null` if"

This adds in particular the ability to track the time at which the
framework boots up, and the time at which we are confident we have
completed the first useful frame.

The old names were getting silly and started stepping on valuable namespace.

The new names are consistent and clear.

Bindings now have a debugRegisterServiceExtensions() method that is
invoked in debug mode (only). (Once we have a profile mode, there'll be
a registerProfileServiceExtensions() method that gets called in that
mode only to register extensions that apply then.)

The BindingBase class provides convenience methods for registering
service extensions that do the equivalent of:

```dart
void extension() { ... }
bool extension([bool enabled]) { ... }
double extension([double extension])  { ... }
Map<String, String> extension([Map<String, String> parameters]) { ... }
```

The BindingBase class also itself registers ext.flutter.reassemble,
which it has call a function on the binding called
reassembleApplication().

The Scheduler binding now exposes the preexisting
ext.flutter.timeDilation.

The Renderer binding now exposes the preexisting ext.flutter.debugPaint.

The Renderer binding hooks reassembleApplication to trigger the
rendering tree to be reprocessed (in particular, to fix up the
optimisation closures).

All the logic from rendering/debug.dart about service extensions is
replaced by the above.

I moved basic_types to foundation.

The FlutterWidgets binding hooks reassembleApplication to trigger the
widget tree to be entirely rebuilt.

Flutter Driver now uses ext.flutter.driver instead of
ext.flutter_driver, and is hooked using the same binding mechanism.
Eventually we'll probably move the logic into the Flutter library so
that you just get it without having to invoke a special method first.

If two repaint boundaries mark themselves dirty, but the second one is a
child of the first, then the second one will get repainted by the first
and then when we come to paint it directly, we get confused because it
isn't dirty any more.

We ran into this in layout before. Apply the same fix. Also, apply the
same fix to composition while we're at it.

We're getting back to the point where we have a bunch of foundation APIs
and it's getting confusing having them mixed with services/.

This also shrinks the width of the error messages a bit because now that
we use 'package:' URLs the stacks are a bit narrower.

I ran into a case where I was setting minHeight=∞ and then calling
layout() with that constraint, which is all kinds of bad. To try to
catch this earlier, this patch now provides a way to catch constraints
that are requiring infinite values.

We don't _always_ check this because there are valid uses for
BoxConstraints.biggest, e.g. as an additionalConstraint.

Fixes #2855

Remove public accessor for paint dirty state

- Create a FlutterErrorDetails struct-like class that describes an

  exception along with more details that aren't in the exception, like

  where it was caught and what was going on when it was caught.



- Provide a FlutterError static API for handling these objects:



  - FlutterError.onError which is called whenever Flutter catches an

    error.



  - FlutterError.reportError() which handles an error.



  - FlutterError.dumpErrorToConsole() which is the default behavior

    for onError.



- Removes all the existing exception handler callbacks.



- Replaces all the existing places that described exceptions using

  debugPrint with calls to FlutterError.reportError().



- Extend lockState() to also catch exceptions, so that we catch

  exceptions that happen during finalizers.



- Make the test framework catch errors and treat them as failures.



- Provide a mechanism to override this behavior in the test framework.



- Make the tests that used to depend on the exception handler

  callbacks use this new mechanism.



- Make pump() also support the phase argument.



- Improve some tests using these new features.



Fixes #2356, #2988, #2985, #2220.

For example, to implement backdrop blur effects.

The performLayout and build callsite are highly megamorphic because they
dispatch into a large number of clients. However, for a given caller, the
callee is always of the same type, which means the megamorphic lookup exactly
factors by the caller. We can speed up the dispatch by capturing a closure at
initialization and then monomorphically dispatching through the closure.

The elaboration will continue until morale improves.

Specific exceptions considered here:

 - size setter checking when you set the size
 - layout verifying that you do set the size
 - hitTest verifying that you aren't dirty and have a size
 - flex complaining about canFlex

We can work around these in code rather than by post-processing the analyzer
output.

Fixes #2940