test(android): command-coverage manifest parity with iOS

[thymikee]

Part of #1412 (Track D).

Context

iOS has a registry-derived manifest classifying all public commands (live / command-contract / workflow-live / capability-denial / known-gap) with static gates (#1408). Android has replay smokes and nightly suites but no equivalent "every command has an owner" matrix; macOS/web/TV are thinner still.

Deliverables

• Generalize the manifest types (they already support the needed row kinds) to a per-platform dimension, starting with Android: one entry per public command on an Android emulator target, satisfies Record<PublicCommand, ...> so a new command fails until classified.
• It is EXPECTED that many rows start as command-contract or known-gap with tracking issues — the point is converting platform parity from vibes into a gated, visible list. Do not write new live scenarios beyond what the existing Android smoke/replay lanes already prove; credit those first.
• Static contract test running on every PR (like smoke-ios-simulator-coverage.test.ts), asserting classification completeness and that named evidence exists.

Acceptance

• Coverage report artifact enumerating live vs contract vs gap counts for Android.
• Every known-gap row carries a tracking issue reference (create them as part of this work).

Out of scope

macOS/web/TV manifests (follow-ups once the pattern is proven twice).

Review amendments (2026-07-27)

• The out-of-scope platforms become tracked checkboxes HERE so deferral cannot silently become permanent. Follow-up ledger (each can start mostly as classified gaps — completeness over live coverage):
  • macOS manifest
  • web manifest
  • Linux manifest
  • tvOS manifest
  • visionOS / watchOS decision row (support or explicit capability-denial)