This concept is now private to AnimationController. All the clients actually
want the AnimationStatus.

Previously we would elide forward and reverse callbacks that canceled each
other out, which broke the expected state machine. Now we synchronously deliver
status callbacks when start an animation.

Fixes #1913

This assert was overzealous. You can update a RawGestureDetector any
time before the semantics phase, it doesn't have to only be during your
child's layout.

This is good because Block has additional PointerListeners between its
GestureDetector and its Viewport.

Now onScroll callbacks are always bracketed by onScrollStart and onScrollEnd
callbacks.

Fixes #1822

Somehow this actually made heroes way simpler.

We don't need these keys. Also, improve style in several places.

Fixes #2225

...and friends.

See the comment in the patch for details.

Refactor Dismissable to not need a SizeObserver and to generally tighten
up the code a bit.

Also:
 - add operator==/hashCode/toString to ViewportDimensions
 - add toString to BindingBase
 - add toString and debugFillDescription to ScrollBehavior
 - fix a bug in the RawGestureDetectorState's replaceGestureRecognizers
 - rename MixedViewport's onExtentsUpdate to onExtentChanged
 - replace ExtentsUpdateCallback with ValueChanged<double>
 - remove a microtask for dispatching scroll start, since it
   did not appear to have any purpose
 - added dartdocs to Instrumentation until I understood it
 - made all event dispatch in Instrumentation drain microtasks

There's no reason to make clients supply a PointerRounter and a
GestureArena when constructing gesture recognizers. These objects are
statics and the gesture recognizers can just grab them directly.

Also, remove the callback constructor arguments. Almost no code used
them. Instead, people seem to prefer using the `..` operator to set
callbacks on the recognizers. Removing the arguments removes a bunch of
boilerplate.

Otherwise, changing the order of your children (e.g. shuffling a list of
children) will have no visible effect.

Currently the interface for recognizing gestures on text spans is pretty ugly,
but hopefully we can improve it with time.

Fixes #156

We forgot to compare the lengths of the lists.

This allows us to adjust exactly which gestures we're listening for
during layout, which I'll use to kill a SizeObserver.

One bug was masking the other, hence they both needed to be fixed
and tested separately.

@Hixie

Now we just have one TextSpan class that handles both simple strings, trees of
children, and styling both. This approach simplifies the interface for most
clients.

This patch also removes StyledText, which was weakly typed and tricky to use
correctly. The replacement is RichText, which is strongly typed and uses
TextSpan.

Rather than clamping _value on read, we now clamp the value when writing it,
which simplifies reasoning about _value.

Turns out we don't need 'parent' outside the assert at all.

This makes ScrollBehavior agnostic about the kind of scroll.
This also means ExtentScrollBehavior's comment about being linear is
more relevant (previously it was true of all scroll behaviours).

In the dark theme, there isn't really a primary swatch, so this API was
a sandtrap. Instead, be explicit about the colors we need for various
widgets in the theme.

Fixes #1277

This helper makes it easier to write correct tests that involve
gestures.

Fixes #1855

These files are already in the "material" folder. They don't need to be
prefixed with material again.

Fixes #702

We should mutate the status of the AnimationController before calling
the status callbacks so that we're prepared to be re-entered.

Fixes #1911

The example in the spec with multiple actions is an anti-example.

Fixes #1876

Downloading these packages from cloud storage simplifies our deployment
story because we can upload to cloud storage automatically from the
buildbot.

This patch also switches the responsibility for downloading the engine
artifacts to update_engine.sh. Centralizing this responsibility ensures
that the packages and the binaries are always in sync.

We don't need this anymore now that we have Mozart. The engine recently
stopped support it also.