fix: enforce integrity contracts and ship Yeoul 0.3.0

[bhyi4]

Summary

Harden the supervised Yeoul workflow without making preregistration mandatory for discussion arcs.

• Recompute every MIRROR-SPEC hash/link, pin the first registration, and reject unsealed/tampered/replaced/deleted bindings. Legacy links require explicit verified migration.
• Freeze TODO criteria before work; share CLI/MCP eligibility; prevent changed commands/item deletion from passing verification. Bound rounds/timeouts and stop on unknown usage.
• Keep drafts/refusals pending, reject missing defense fields and archive collisions, separate archive recording state, and show active arcs before historical verdicts.
• Add an explicit supervisor-driven result adapter for Mirror's bound result contract; it does not publish or infer a scientific verdict.
• Bundle scripts/templates/license in wheels and sdists, test the installed package outside the checkout, and refresh docs, command references and migration guidance.

Verification

• Existing gate suite: 72/72.
• New adversarial contract suite: 26 tests, including relocated defense fields and preserved prior-run logs.
• MCP subprocess contract: 8/8.
• MCP serverInfo wire contract: 8/8 on the host (sandbox pipe did not answer).
• CLI/MCP parity and real Mirror interoperability: 5 tests, including pass/fail/inconclusive result cases and Windows Git Bash discovery.
• Lifecycle demo and pre-publish checks pass; relative documentation links and six public URLs checked.
• sdist -> wheel built and installed harness exercised away from source. CI now checks the packaged paths too.

Compatibility and honest scope

v0.3.0 intentionally requires a supervisor baseline for protected standalone verification; --current-only is an explicit weaker diagnostic. Existing ledgers/archives are not rewritten. Local hashes do not establish identity, external time, content truth or independent reproduction. Agents with permission to rewrite every artifact remain outside this workflow-level boundary. Token limits are between-round controls, not hard provider spending caps.