docs: rewrite the README around what the tray actually is

[abrichr]

Part of the README pass across the public repos
(openadapt-flow#423
has the diagnosis).

Thirty lines went by before this file mentioned a tray. It opened with the
"OpenAdapt is a governed demonstration compiler" paragraph that appears word
for word in five other repositories, and the install command was at line 165.

The opening now says the thing the rest of the file spends its caveats
circling: this is a status surface. It records nothing, compiles nothing,
replays nothing. Saying that in the first sentence makes the whole document
shorter, because half the hedging existed to walk back an implication the
opening created.

"Release boundary" and "Known gaps" were two sections hedging the same three
facts. They're one section now, and it's more direct than either was.

Verified against 0.3.3 from PyPI

Two things the old README got wrong or missed:

• The wheel ships a second entry point, openadapt-tray-gui. It was
  undocumented. It's in the install block now. Both entry points call
  openadapt_tray.app:main; the only difference is that this one is registered
  under gui_scripts, so Windows launches it without a console window. It is
  not a different program.
• The config sample listed "stop_recording": "<ctrl>+<ctrl>+<ctrl>" with no
  explanation, which reads like a typo. It isn't: stop_on_triple_ctrl on this
  package's TrayConfig in config.py defaults to True, and that's the
  hotkey it enables. I kept the line and said so.

Also checked the HotkeyConfig field names and deployment_lane against the
installed package rather than the source tree.

280 lines to 176. Zero em dashes.

The three README strings a test pinned

tests/test_public_metadata.py matched three literal strings against headings
this rewrite removed. All three facts are still in the file, and two of them
moved earlier:

Test asserted	Where it is now
Lifecycle: Experimental supporting surface	"this is a status surface, not an integrated desktop product", second paragraph
Release boundary (heading)	"It records nothing, compiles nothing, and replays nothing", first paragraph
not been validated together end to end	unchanged, first bullet of What this doesn't do yet

The assertions now name the claims instead of the headings, and they collapse
whitespace before matching, so a line rewrap can't break an assertion on a
sentence a reader would still see. Every negative assertion is untouched:
openadapt train, Training Control and monitoring training must still not
appear.

Open question for a reviewer. If you want the literal
Lifecycle: Experimental supporting surface string back in the README, say so
and it goes back. The argument for dropping it: the registry at
OpenAdaptAI/.github/REPOSITORY_LIFECYCLE.md is the canonical place for a
lifecycle label, and a plain sentence about what the thing does not do is harder
to misread than a label is.

Two sibling repositories carry the same trap. openadapt-flow pins three
licensing strings in tests/test_openimis_claims_fixture.py, and
openadapt-capture pins an admission-state claim in
tests/test_public_lifecycle.py. Grep tests/ for the file name before
rewriting a README anywhere in the org.