[Refactor] Consolidate delegated model consent on canonical dialogue

[roomote-roomote]

​Created by Roomote. View the session or mention @roomote-roomote for follow-up asks.

Related issue

Maintenance cleanup of the canonical model-request dialogue introduced in #3439; no linked issue.

Why this PR exists

• A maintainer explicitly invited this PR in the linked issue or discussion
• I am a maintainer / this is internal Roomote work

What changed

• Require canonical typed dialogue as the model resolver's evidence input. Remove the parallel human-text/proposal contract, synthetic dialogue fallback, unused proposal selector, and redundant derivation in both task-launch and PR-review callers.
• Remove the test-only model-proposal recovery mode from ordinary pre-prompt reply retrieval, preserving its existing live query behavior.
• Migrate resolver regressions to typed dialogue and cover stale-proposal human boundaries through both canonical and compacted replay dialogue. Retain canonical provenance, equal-timestamp insertion boundaries, queued confirmation/revocation, durable snapshots, and atomic snapshot preservation coverage.

Both production resolver callers already supply dialogue. This behavior-preserving simplification removes competing evidence contracts without changing authorization policy, classifier thresholds, routing precedence, or snapshot persistence. Net deletion: 85 lines across six files.

How it was tested

• Focused launch-model, model-authorization, conversation-repository (real test database), and service suites: 526 tests passed across the four files after correcting the new parameterized test's table shape and rerunning the affected launch-model suite.
• Cloud-agents typecheck passed; pre-push oxlint, residual ESLint, workspace fast typechecks, and knip passed.
• Browser proof is not applicable to this internal contract deletion; call-site inspection and the canonical dialogue/replay tests provide the relevant evidence.
• The pre-commit semantic judgment hook completed without blocking but left three prose criteria incomplete because the launch-model test exceeded its file evidence budget; a full semantic check remains required before merge.

Checklist

• The PR title follows the repo convention: [Fix], [Feat], [Improve], [Refactor], [Docs], or [Chore] followed by a user-facing description
• This PR is small and scoped to one change
• pnpm lint and pnpm check-types pass locally
  • Fast equivalents passed via the pre-push gates; formatting-inclusive full commands were not run.
• I added tests or included a clear manual validation note above
• I removed secrets, tokens, private keys, and customer data from code, logs, and screenshots
• If this change should appear in the changelog, I ran pnpm changeset
  • Internal behavior-preserving cleanup; no changeset needed.

[roomote-community]

No code issues found. See task

Reviewed all six changed files, both production resolver callers, canonical consent boundaries, replay snapshots, queued follow-ups, and the remaining reply-retrieval caller. The removed fallback and proposal-recovery mode have no production consumers; the remaining query behavior is unchanged.

Current-head CI: Test, Lint, Type Check, Knip, and CodeQL passed. Docker Build (app, amd64) is still running.

_{Reviewed 518727b}