[Live Caption] Implement "non-web" prototype.

This CL implements a prototype of live captioning for non-web (e.g.
Android apps, linux) audio sources.

The prototype is gated behind the feature flag 'SystemLiveCaption'
because it is currently deficient in the following ways:
  1) It uses the input device stream instead of a bespoke "non-web"
     loopback. This means that captions are generated for speech
     detected through the microphone. The required loopback doesn't yet
     exist.
  2) It is always active, c.f. running only when some non-web audio
     stream is active. This is a huge drain on battery and processing
     power. The cras interface indexing non-web sound producers also
     doesn't yet exist.

Note that tests for this CL live in a subsequent CL (that depends on
this one), since a a much broader footprint is required to expose
information for testing purposes.

Bug: b:253114860
Change-Id: I52152ee070e3abd4d6b1772d9523f4311ccabb25
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/4050012
Reviewed-by: Toni Barzic <tbarzic@chromium.org>
Reviewed-by: Evan Liu <evliu@google.com>
Commit-Queue: Michael Martis <martis@chromium.org>
Reviewed-by: Andrew Moylan <amoylan@chromium.org>
Reviewed-by: Katie Dektar <katie@chromium.org>
Cr-Commit-Position: refs/heads/main@{#1080016}