Recover v0.9 execution evidence and add durable fail-stop supervision

[DavidBakerEffendi]

The v0.9.0 run stopped at its eighth control and later lost temporary execution directories. Preserve surviving evidence, register a disabled recovery candidate on durable paths, and catch malformed commands before starting analyzers.

Key Changes:

• Preserve historical identities and seven transcript completions as unqualified; record byte-exact restoration of the Swift extractor, repaired packs and stock runtime assets.
• Add bounded, durable command supervision without retries and parser checks for all 117 prospective commands (93 Rust, 19 shell, five Python). Rust validation returns before population setup or analyzer dispatch, with missing-argument, invalid-mode, unknown-flag and help-bypass regressions.
• Relocate the prospective CodeQL plan while leaving historical plans and scripts intact. Execution stays disabled pending qualification receipt, merged harness binding, global attempt accounting and fresh reservation.

Touch Points:

• recovery/2026-09-30-execution-loss/
• reports/releases/v0.9.0/execution-v1/recovery-20260930-01/
• scripts/supervise-release-command-v090.py and parser/supervisor tests
• src/main.rs, src/cli_args_tests.rs, .github/workflows/ci.yml

Validation: local shell/Python parser tests (five tests covering 24 commands), six subprocess supervisor tests, formatting, historical v0.8 planning check and canonical execution-contract check pass. Hosted CI validates the Rust commands and full repository gates. Local Cargo deferred for active Bifrost work. No benchmark rerun; supervisor process-group signals do not prove containment of descendants that create new sessions.

Scope: parser coverage verifies syntax, required arguments and parser-defined enums only. Population/runtime setup and dispatch-time cross-field checks are intentionally skipped; all final commands still require real launch/preflight qualification.