The right fix here is to use a radial reaction, but for now we can just make
the gesture detector including the padding acround the icon in its hit test
region.

Fixes #575

Now a RenderBox is considered hit if one of its children are hit or it itself
decides that it's hit. In particular, empty space inside a flex won't be hit
because none of the children are located there and a RenderFlex doesn't
consider itself hittable.

Fixes #53
Fixes #1221

The name of the parameter wasn't consistent with IconButton or DrawerItem, etc.

Fixes #1871

Material design icons are defined to work at specific sizes: 18, 24, 36, 48.
The current API doesn't reflect that and just takes a size int. If an invalid
size is chosen an error is printed to the console and no icon shows up.

Fixes #1816

Previous these callbacks were leaking the implementation detail that they were
triggered by taps. In a later patch, we're going to add a parameter to
GestureTapCallback that these callbacks won't have.

Related to #1807

Provide a way to schedule a callback for immediately after the current
frame has been sent to the GPU thread. This is useful for times where
you want to immediately schedule yourself for another build or layout
because, for instance, you just displayed frame zero of an animation and
you want to use the metrics from that layout to set up the actual
animation to begin immediately, such that the very next frame is frame 1.

Instead of requiring clients to use ui.ColorFilter, we show show the
ColorFilter class, similar to what we do for Color.

A MaterialList understands the sizing, padding, and scrollbar features of
Material Design lists.

Also, add CircleAvatar for showing the circular avatars that are commonly used
in material lists.

- truncate pixel values to 1dp since there's really no point being told
  the Size is 302.98732587287 by 648.28498579187.

- describe more Widgets so that debugDumpApp() is more useful.

- remove bufferValue from ProgressIndicator (cc @hansmuller) since it's
  not yet implemented.

- half-hearted toString() for ThemeData. There's no point making a
  complete one, since it would cause line-wrap even on big monitors in
  debugDumpApp dumps, and you can easily get the actual values from a
  debugging if that's the issue.

- flesh out BoxConstraints.toString() so that fully unconstrained and
  fully infinite constraints are called out explicitly. I experimented
  with adding even more special cases, e.g. calling out unconstrained
  widths with fixed heights, etc, but it made the output less readable.

- remove a redundant _updateVariable() in AnimatedContainer (cc
  @abarth).

- add more information to RenderView.toString().

The goal is to follow the guidelines in
https://github.com/flutter/engine/blob/master/sky/specs/style-guide.md#packages

Fixes #1638

Instead of using a ColorFilter, plumb the information down to Icon and thereby
to Image.

These are now part of material.dart.

After this patch, InkWell is driven by gesture recognizers, which lets us
cleanly cancel splashes when the user actually scrolls.

I've also refactored all the clients of InkWell to use InkWell to detect
gestures instead of wrapping InkWell in a GestureDetector.

Fixes #1271

Code outside of package:sky should import this code using

package:sky/rendering.dart
package:sky/widgets.dart

Moving this code into the "src" directory is a convention that signifies that
and it cleans up the generated dartdoc because the libraries in the src
directory aren't included in the generated documentation. Instead, the classes
are documented in the widgets.dart and rendering.dart libraries.

We're now using it at the widget layer for everything except scrolling and
flinging.

These gestures use the new gesture detection system.

Sadly, box.dart has grown much longer than 1000 lines. This patch splits it up
into several files based on the class hierarchy. Fortunately, many of these
classes are loosely coupled to each other.

It was confusing to have both widget.dart and widgets.dart

In theory, before, if you had the same image twice in a scrolling container, you'd get an assertion with no way around it.
This makes those nodes not bother making keys by default, which is cheaper and more correct.

This fixes some theoretical bugs whereby we were using hashCode to try
to get unique keys for objects, but really we wanted object identity.
It also lays the groundwork for a new GlobalKey concept.

I tried to keep the impact on the code minimal, which is why the "Key"
constructor is actually a factory that returns a StringKey. The code
has this class hierarchy:

```
   KeyBase
    |
   Key--------------+---------------+
    |               |               |
   StringKey    ObjectKey       UniqueKey
```

...where the constructors are Key and Key.stringify (StringKey),
Key.fromObjectIdentity (ObjectKey), and Key.unique (UniqueKey).

We could instead of factory methods use regular constructors with the
following hierarchy:

```
   KeyBase
    |
   LocalKey---------+---------------+
    |               |               |
   Key      ObjectIdentityKey   UniqueKey
```

...with constructors Key, Key.stringify, ObjectIdentityKey, and
UniqueKey, but I felt that that was maybe a more confusing hierarchy.
I don't have a strong opinion on this.