feat: Codex background process evidence and original-start chronology (v1.22.0)

[alloevil]

Scope

• Link explicit exec_command process IDs to unique later write_stdin results, preserving launch/poll/exit source references.
• Keep reused IDs, duplicate/overlapping polls and contradictory outcomes unknown; count unlinked polls.
• Refine recognized check completion without moving the original launch index. Compound/input-fed check outcomes remain unknown.
• Separate process outcomes from unchanged historical health, failure/recovery and review fingerprints.
• Add live synthetic and hosted Codex examples; version/lockfile/claims synchronized for v1.22.0.

Evidence

• 313 Node tests including 23 process cases and 40 deterministic terminal transformations pass.
• Normal/demo builds, lint (existing findings unchanged), actionlint and 16 automated claims pass.
• Frozen 15-session regression: exact prior health/diagnostics unchanged. 60 Codex starts and 53 polls; all 53 polls uniquely linked, with 48 successful exits, 1 failed exit, 11 last-recorded running processes.
• Browser: launch/final-result jumps, overlap semantics, later-check failure, unlinked polling, live completion, 390px layout and static demo zero API calls pass.
• Generated fixtures and private regression outputs reproduce byte-for-byte. Staged secret scan passes; no private log IDs/commands/artifacts included.

Limits

Not live job control, not cross-session or cross-platform lifecycle inference, not task/test coverage proof. Publish through existing Release workflow only after protected checks.