refactor(information-architecture): migrate mechanics to Python

[BjRo]

Why

Closes #180. Move information-architecture mechanics off Bash so setup and doctor can run through one contained Python + UV package on POSIX and native Windows.

What changed

• Replace the setup and doctor Bash entrypoints with frozen ia-setup and ia-doctor commands, preserving inventory records, runtime selection, routing, reachability, adapters, deduplication, budgets, and refusal statuses, and UTF-8 path output.
• Add ia-write for approved atomic guidance updates with expected-content checks, preserved newline/BOM conventions and permissions, and symlink-source handling.
• Add native regression/property tests, runtime-only copied-artifact validation, and Linux/macOS/Windows CI; register the package, update both skills and eval fixtures, and bump both manifests to 0.2.1.
• Port the remaining legacy shell regressions into pytest, strengthen existing native assertions, and remove both Bash suites and their CI steps. All regression cases now run through the cross-platform Python quality matrix.

Verification

• bun run check:python: passed across all registered packages; 104 target tests, 99.67% statement coverage, 98.44% branch coverage, strict mypy and Ruff.
• The original Bash suites passed before removal. Their regression scenarios are now covered by native pytest tests; backend/tests/acceptance.md maps the retained coverage. The IA plugin contains no Bash tests or runtime launchers.
• uv run --quiet --frozen --no-dev --project plugins/foundation/darrow-information-architecture/backend python plugins/foundation/darrow-information-architecture/backend/tests/fresh_install.py: passed locally and in CI on Linux, macOS, and native Windows with a fresh independent plugin and runtime dependencies only.
• Both bundled skill inspections and claude plugin validate plugins/foundation/darrow-information-architecture: passed.
• bun run lint, bun run lint:ts, bun run lint:shell, bun run typecheck, bun run check:decisions, bun run check:docs, bun run check:docs:external, bun run check:python:inventory, and scripts/test-python-quality-changes: passed.
• bun test: 620 passed, two failed. The recovery timeout passed in isolation; the remaining pre-existing discovery test expects a shell check that is now in semantic_output_checks (evals/runner/discovery-eval-checks.test.ts:104). Neither discovery file is changed here.
• Both IA skill suites: all 23 cases passed task and activation checks on Codex (gpt-5.6-terra, medium) and Claude (claude-sonnet-5, medium), one fresh passing trial per case/host with an 80% case threshold. The final procedure-routing guidance also passed a targeted Codex rerun. Earlier diagnostic failures and corrections are retained; these runs do not establish repeated-trial stability.
• Python CI: all IA Python 3.10–3.13 jobs passed on Linux, macOS, and Windows at aa2951c, together with all three fresh-install jobs. Documentation CI also passed.

Review notes

The old Bash paths are removed; callers use the frozen package commands. The audit remains a compact structural check, not a complete Markdown or runtime-loader parser. Atomic replacement detects observed source changes but requires coordination with concurrent writers and is not a multi-file transaction.

The independent review's parent-relative resource and unselected-runtime encoding findings are fixed and covered by regressions. Eval calibration changes and the procedure-routing correction are recorded in backend/tests/acceptance.md; full pre-migration live skill baselines are not claimed.

Checklist

• I have read and followed CONTRIBUTING.md, including the contribution
  licensing terms.
• I added or updated the applicable invariant before implementation, or
  this change does not affect a capability invariant.
• I added or updated colocated evals, or this change does not affect skill
  behavior.
• I confirmed that each changed plugin remains self-contained, or this
  change does not affect plugin content.
• I ran bun run check:python, or this change does not affect registered
  Python packages or their repository quality infrastructure.