refactor(adaptive-delivery): migrate mechanics to Python

[BjRo]

Why

Adaptive-delivery's preflight, host routing, and evaluation fixtures relied on Bash/AWK mechanics that prevented native Windows use. Closes #183.

What changed

• Replace substantive shell mechanics with a self-contained Python 3.10–3.13 package, frozen UV entrypoints, and no runtime dependencies. Preserve public records, route validation, repository binding, refusal behavior, and the host-owned single-owner orchestration contract.
• Migrate readiness, review, verification, and completion-proof fixtures and their callers; retain shell regression scenarios as development checks. Keep review-only fixture tokens confined to their advertised skill, and keep legacy Codex shell attestations separate from the Python package.
• Register Python quality gates and native Linux/macOS/Windows fresh-install coverage, update the specification and invocation docs, and bump both plugin manifests to 0.22.1. Clarify Claude's existing explicit foreground-launch requirement.

Verification

• bun run check:python — passes, including Ruff, strict mypy, package tests, and separate 95% statement/branch gates.
• Native Windows fresh installation exposed legacy-code-page output; a regression reproduces it with PYTHONIOENCODING=cp1252 and now passes with explicit UTF-8/LF console output. Final local package coverage is 99.87% statements and 99.28% branches.
• bun run lint, bun run lint:ts, bun run lint:shell, bun run typecheck, bun run check:decisions, and bun run check:docs — pass.
• Six shell regression suites pass under both bash and /bin/bash (3.2). Python CI passes, including Python 3.10–3.13 on Linux/macOS/Windows, fresh copied runtime-only installation on all three platforms, and Bash 3.2/5 regression scenarios.
• bun run check:docs:external — passes (86 external URLs).
• Codex adapter tests: 60 passed; readiness/review fixture tests: 34 passed. The full runner suite returned 622 passed and one unrelated discovery test failure (planning transfer does not turn missing repository policy into authority), reproduced on unchanged base commit 50d796e.
• All 55 adaptive-delivery eval fixtures prepare on both hosts. Live mechanical owner case: Codex 1/1; Claude final 5/5. The initial Claude sample was 2/5 because three launches omitted the explicit foreground flag; the existing requirement was clarified and the same case rerun without assertion changes. Matched original-plugin Claude control: 1/1.
• Independent Standards and Spec reviews identified three blockers; targeted repairs and independent fix verification are retained locally. Full live evaluation of every orchestration case was not run.

Review notes

The public commands now use uv run --quiet --frozen --no-dev --project <absolute-plugin-backend>; obsolete Bash entrypoints are removed. Invalid JSON syntax uses standard-library line/column diagnostics; semantic validation and exit/status contracts are preserved. The unrelated discovery test remains unchanged; all PR CI checks pass.

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.