Refs #94.

Looking at Google Play's migration documentation [1] it seems we
might get away with bumping the target sdk level and not changing
anything else \o/

[1] https://developer.android.com/about/versions/13/behavior-changes-13

... that was added as part of the Obelisk scaffolding.

We can avoid carrying the three-tuple of result values over the scrim div by
composing the container div and the scim div with `<*` which sequences the two
actions but then discards the result of the right-hand side and propagates the
result of the left-hand side (where the tuple is produced).

Refs #89

Optional in sdk 31, defaulting to true, but required by Google Play Store for
uploaded bundles.

This changes us to use a fork of reflex-platform with two useful features:

* It accepts arguments controlling the Android sdk versions written to the
  Gradle build file.  We use this feature to bump our target sdk version.
  This will let us build bundles accepted by Google Play Store.

* It has a fix to the Nix expression for Android "Bundle" builds that makes
  the install phase succeed by copying the correct file around.

The latter is not directly relevant to sdk 31 support but it's in the
integration branch so by switching to that branch we're getting it.  This is
great because we'll want it in a minute anyway.

Android sdk 30 changes the default behavior for allow access to files from the
webview component from true to false (supposedly for security-related
reasons).

There is a reflex-dom branch (dylang/android_fixes_dev) includes a fix for
this which is cherry-picked into the revision now pinned by our reflex-dom
thunk.

We do not update directly to that branch since it is based on a fairly new
reflex-dom which comes with a cascade of dependency version updates which are
challenging (at least) to resolve.

With this change, static assets now load properly on Android sdk 29 (what we
currently use), Android sdk 30 (where the Android behavior changed) and
Android sdk 31 (the minimum version required for app uploads to Play Store
now).

Not to mention be simpler.
Maybe we don't even want/need nix-optics now ... ?

To do this, take over direct management of the reflex-platform dependency.