* MultiTapGestureRecognizer previously would assert if there was no
   competition.
 * GestureArenaTeam would always select the first recongizer as the
   winner even if a later recognizer actually accepted the pointer
   sequence.
 * debugPrintStack would fail a type check if maxFrames was non-null.
 * FractionalOffset.lerp would throw a null-pointer exception if its
   second argument was null.

Also, add a number of tests for previously untested lines of code.

Previously, the Slider used a drag gesture recognizer to move the head
of the slider, but when the slider was in a vertical scroller, the
recognizer would wait until the user moved the pointer by enough pixels
to disambiguate between sliding the slider and scrolling the scroller.

That worked fine for actual drags, but the slider should also move when
the user taps the track. This patch introduces a tap recognizer to
handle that behavior.

To avoid the slider's drag and tap recognizers from competing with each
other in the arena, this patch introduces the notion of a
GestureArenaTeam, which lets several recognizers combine to form one
entry in the arena.  If that entry wins, the team picks the first of its
recognizers as the winner, avoiding latency.

Fixes #7454

This patch improves test coverage for a number of files that had low
coverage. This patch also fixes a few minor bugs found by these tests.

- more dartdocs for the drag typedefs

- more toStrings to aid debugging

- require the position for DragUpdateDetails since we were omitting it
  in some places

- add the primaryVelocity to DragEndDetails so that consumers don't
  have to themselves track the axis in question

- fix the velocity tracker so that it doesn't walk the null data.
  Previously, near time t=0 (which pretty much only matters in tests,
  but it does matter there) we would walk the velocity data and then
  also walk missing data, treating it as Point.zero with t=0.

- simplify some of the velocity tracker; e.g. instead of trying (and
  failing?) to clear the velocity tracker when the pointer stalls,
  just drop the data before a stall during the velocity estimation
  (where we redundantly had another bigger horizon anyway).

I used the wrong identifiers here.

See https://github.com/flutter/flutter/pull/6861

This silences all but two of the warnings from https://travis-ci.org/flutter/flutter/builds/176055550

One of the silenced warnings was a genuine code error.

Some of the others were correct but there was no way for the analyzer to know, and I worked around them.

The remainder are problems with the analyzer, specifically https://github.com/dart-lang/sdk/issues/27827 and https://github.com/dart-lang/sdk/issues/27504.

If a recognizer is interrupted by an exception from a callback, it could be
left in an inconsistent state and be unable to process future events

Previously, we were nulling out the ArenaEntry in MultiDragPointerState
too early, which was prevent MultiDragPointerState from rejecting the
gesture in `dispose` if we hadn't accepted by the time the pointer went
up. Now we properly reject the gesture, which causes the tap gesture to
win during `sweep` in the arena.

Also, add a bunch of docs and annotations based on information I learned
while studying this issue. Finally, rename a private member of tap
recognizer to a name that would have confused me less in my
investigation.

Fixes #1186

Previously we would maroon the feedback in the overlay. Now we let the
drag proceed and only tear down the gesture recognizer when all the
drags it spawns have been resolved.

Fixes #6151

Previously we were relying on the gesture arena to call us back to
cancel our timer. However, in the case where we've already been
accepted, asking the gesture arena to reject us doesn't lead to a
callback and we fail to stop the timer (and hence trigger an assert).

Fixes #6156

Now dart:ui does the decoding of the pointer data itself, which means we don't
need to do it in the framework.

This rewrites imports of various mojom.dart files from the Flutter
engine repo to instead import normal-looking dart files from the
(new) flutter_services package. This package handles exporting the
correct symbols from generated code wherever that may live.

Includes an engine roll to 3551e7a48e2e336777b15c7637af92fd7605b6c5
which contains the new flutter_services package.

The engine now reports coordinates in physical pixels.

I did a pass through some of the code cleaning minor things up.

Fixes #4761

This patch updates the framework to use the new SemanticAction in
semantics.mojom.

This patch adds tests for some code paths we weren't hitting before and removes
some dead code that couldn't be tested because it was unreachable.

* Change how navigator prevents redundant operations

Instead of requiring transactions, we now cancel all active pointers that are
interacting with the navigator and absorb future pointers until we get a chance
to build. This approach isn't perfect (e.g., events that trigger off the
cancelled pointers could still interact with the navigator), but it should be
better than the current transaction-based approach.

Fixes #4716

* Remove openTransaction

* test

* fixup

In writing a test for #1927, I found a number of bugs in how Draggable shuts
down. Previously it would leak its recongizer. Now it disposes its recognizer
and the recognizer knows how to be disposed cleanly.

Fixes #1927

Fixes #4428

We have these details objects for the same reason we now have drag details
objects: future extensibility.

We were trying to unregister the pointer route twice. Now we only unregister it
once.

Fixes #4341

I changed the behavior late in the code review but didn't update this dartdoc.

This patch starts working through the missing dartdocs in widgets.dart.

Previously we supplied individual parameters to the various drag and pan
callbacks. However, that approach isn't extensible because each new
parameter is a breaking change to the API.

This patch makes a one-time breaking change to the API to provide a
"details" object that we can extend over time as we need to expose more
information. The first planned extension is adding enough information to
accurately produce an overscroll glow on Android.

This patch starts using the mustCallSuper annotation for several of the key
lifecycle callbacks in the framework. I haven't added it to didUpdateConfig
because there are a large number of non-compliant overrides. We should discuss
whether we want to use it there.

We'll use this functionality to implement some subtle behaviors for tooltips.

For consistency.

Fixes #4142

Also, make some previously private classes public for better
documentation.

Also, add some missing docs to http.dart and widgets.dart.

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`.