This is an initial implementation of Select-to-speak. Given mouse events
that are "captured" - currently by holding down Search - it draws a box
around the area the user selects, and then speaks everything within that
box based on its accessible name.

This is not meant to be a final, complete implementation, just a good
starting point to begin experimenting.

Depends on: https://codereview.chromium.org/2493923002/

BUG=593887
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:closure_compilation

Review-Url: https://codereview.chromium.org/2509883002
Cr-Commit-Position: refs/heads/master@{#435555}