fix(evolve): recover AI progress across navigation

[Xnhyacinth]

Summary

This PR stabilizes the AI/Evolve and Twin flows after the latest main merge. It focuses on restoring backend-backed results, keeping active AI runs visible after navigation, and hardening Twin run recovery/sync behavior.

Feature Additions

• Add Twin recent run history support so users can see and resume the latest 5 Twin runs from the progress area.
• Add /api/twin/runs and related run metadata helpers for history rendering and exact run resume.
• Add run-scoped Twin reads for overview/cards/traits/events/runtime-preview/avatar/sync so the UI can bind to the active run_id instead of mixing data across runs.
• Add Evolve legacy cache fallback/migration so existing file-cache results for profile, memory, and other tabs can be surfaced and moved into the shared cache path.
• Add Evolve/Twin frontend regression coverage for streaming recovery, cache recovery, navigation state, and run-scoped reads.
• Add AI engine health checks and explicit engine fallback behavior to make Codex/Claude selection more predictable.

Bug Fixes

• Fix #ai profile/memory panels showing empty/no-cache states even when backend or legacy cache results already exist.
• Fix #ai active Evolve runs being hidden by stale local Timeout/network error cache after switching away and returning.
• Fix Evolve server-cache reload paths overwriting active running UI while backend progress is still alive.
• Fix inactive Evolve tab panels being created visible during background preload.
• Fix server restart process matching by normalizing pid values before inspecting command lines.
• Fix Twin progress/resume synchronization after tab reloads or disconnected streams, including exact run_id resume.
• Fix Twin sync safety so CHATVIEW_CLAUDE_MD/test-scoped writes go to the requested target instead of touching the real global ~/.claude/CLAUDE.md.
• Fix Python 3.9 CI imports by replacing Python 3.10+ union annotations in touched modules with typing.Optional / typing.Union.
• Merge latest origin/main and resolve conflicts with main behavior as the base, then reapply the Evolve/Twin recovery logic on top.

Validation

• python3 -m unittest tests.test_api.TestTwinSyncScoping tests.test_api.TestTwinRunScopingReads tests.test_api.TestTwinResumeLatestTrait tests.test_api.TestTwinAvatarScoping tests.test_api.TestTwinResumeCheckpointOnly -> Ran 13 tests ... OK
• python3 -m py_compile server.py analyze.py db.py chatview/ai_engine.py chatview/cli.py chatview/server.py chatview/handlers/twin.py chatview/utils/sync.py -> exit code 0
• python3 - <<'PY' ... ast.parse(..., feature_version=(3, 9)) ... PY -> chatview/utils/sync.py: py39 ast ok, chatview/ai_engine.py: py39 ast ok
• node --check static/evolve.js && node --check static/twin.js -> exit code 0
• ruff check . -> All checks passed!
• python3 -m unittest discover -s tests -> Ran 159 tests ... OK (skipped=3)
• GitHub Actions run 28465262479 -> lint, test (3.9), test (3.11), and test (3.13) passed