fix(v2): show revoked state on mobile rows - #1712

…side (TASK-050)

The UI hold asks for a before/after pair at 390 and 1200. Wren's read named the one
view that can carry this fix on a phone: the row's `… may use it` line is hidden under
760px by v2.css (#1712), so the audience only paints in the Manage aside.

- before = main dbd5270, after = dbd5270 + this PR's head ede0647
- one backend at main, both frontends pointed at it, same seeded pod and grant
- harness: scripts/ui-evidence-shot.mjs on kai/harness-width-flag @ a730a1e (#1770).
  It now takes --width/--height and --click, refuses a flag it does not implement, and
  prints the viewport it used — so these files are 780x1800 and 2400x1800, not the
  2880x1800 that `--width 390` used to produce under a `-390` name.

Measured: aside @390 `agents allowed` an agent -> localdev; row @1200 same change; row
@390 unchanged, which is what the hide rule requires.

…side (TASK-050)

The UI hold asks for a before/after pair at 390 and 1200. Wren's read named the one
view that can carry this fix on a phone: the row's `… may use it` line is hidden under
760px by v2.css (#1712), so the audience only paints in the Manage aside.

- before = main dbd5270, after = dbd5270 + this PR's head ede0647
- one backend at main, both frontends pointed at it, same seeded pod and grant
- harness: scripts/ui-evidence-shot.mjs on kai/harness-width-flag @ a730a1e (#1770).
  It now takes --width/--height and --click, refuses a flag it does not implement, and
  prints the viewport it used — so these files are 780x1800 and 2400x1800, not the
  2880x1800 that `--width 390` used to produce under a `-390` name.

Measured: aside @390 `agents allowed` an agent -> localdev; row @1200 same change; row
@390 unchanged, which is what the hide rule requires.

… agent" (TASK-050) (#1757)

* fix(v2): name a human in a grant audience instead of calling them "an agent"

A pod grant's audience is a snapshot of the room's members, so it can
name a human; `seats` holds agents only, and every id it could not find
was labelled "an agent". The stranger's Tools row read "an agent, an
agent, C4 smoke may use it" for two people.

Resolve the member's username before falling back to the generic label,
the same helper the grantor line already uses. The row named two options;
this is the first one, and the reason is on the row: enforcement needs an
agent identity, so naming the person is accurate under either reading of
what an audience means, and the audience-model question is left to the
pod rather than settled here.

Mutation-proven: dropping the fallback reds exactly the new case.
Frontend 108 suites / 884 tests, tsc clean. TASK-050.

* docs(evidence): the TASK-050 audience pair at 390 and 1200, row and aside (TASK-050)

The UI hold asks for a before/after pair at 390 and 1200. Wren's read named the one
view that can carry this fix on a phone: the row's `… may use it` line is hidden under
760px by v2.css (#1712), so the audience only paints in the Manage aside.

- before = main dbd5270, after = dbd5270 + this PR's head ede0647
- one backend at main, both frontends pointed at it, same seeded pod and grant
- harness: scripts/ui-evidence-shot.mjs on kai/harness-width-flag @ a730a1e (#1770).
  It now takes --width/--height and --click, refuses a flag it does not implement, and
  prints the viewport it used — so these files are 780x1800 and 2400x1800, not the
  2880x1800 that `--width 390` used to produce under a `-390` name.

Measured: aside @390 `agents allowed` an agent -> localdev; row @1200 same change; row
@390 unchanged, which is what the hide rule requires.