Complete 0.5 recovery and 0.6 usage milestones

[lifrary]

Summary

• Reprioritize and complete the roadmap through 0.6.0; keep npm publication separate (latest published remains 0.4.0).
• Preserve bounded, redacted historical instructions, assistant statements and recorded compaction summaries, with explicit continue/fork cutoffs and current-versus-historical Git state.
• Search archived Codex and Claude/Codex child transcripts, exposing exact file provenance and only uniquely proven parent-step links.
• Account for native Codex descendants separately, deduplicate evidenced inherited calls, surface unattributed children, and invalidate MCP caches on descendant changes.
• Add opt-in CLI/MCP session-catalog token totals after applying the catalog limit; preserve cheap default catalogs and distinguish missing data from logged zeros.
• Update public contracts, release metadata, committed bundles and installed-tarball CLI/MCP smoke coverage.

Verification

• Linux server, Node 24.21.0, two CPUs: lint, native and legacy TypeScript checks, 40 test files (956 passed, one pre-existing root-only permission test skipped), build, release consistency, installed-tarball smoke, and npm audit all passed.
• 60 additional attribution/discovery regressions cover ambiguous IDs, optional filesystem failures, archive duplicates, native-path validation and directory replacement races.
• Integrated synthetic recovery benchmark: 2,000 / 8,000 / 24,000 events built in 270 / 906 / 2,617 ms respectively; no timing threshold added to tests.
• Recovery omissions remain explicit; missing response/thread/timestamp evidence is not presented as complete billing verification.

No npm publication or release tag is included.
The macOS canonical-path cache regression is fixed. Focused macOS catalog (11 cases) and attribution (60 cases) tests pass; the complete server release gate chain was rerun successfully.