[codex] Add agent loop eval smoke suite

[mimeding]

Summary

Adds a model-free AgentLoop eval domain and smoke suite for todo, complete, and clarify helper contracts.

This PR is intentionally stacked on top of #986, which already includes #985 beneath it. #985 fixes the current origin/main FluidAudio/PocketTTS build break, and #986 carries the currently green development-plan and CI-readiness base. Once #985 and #986 merge, this PR should reduce to the eval smoke-suite commit.

Changes

• Add agent_loop eval expectations and runner path for pure-data cases.
• Add smoke fixtures for todo checklist parsing, complete summary validation, and clarify option parsing/deduping.
• Skip preflight plugin/model setup for pure-data suites so they run without a live model/server.
• Document AgentLoop suite usage in Packages/OsaurusEvals/README.md.

Validation

• git diff --check
• jq empty Packages/OsaurusEvals/Suites/AgentLoop/*.json
• make evals EVALS_SUITE=Packages/OsaurusEvals/Suites/AgentLoop
• swift build --package-path Packages/OsaurusCore

Dependency

Depends on #986, which depends on #985.

[mimeding]

Author-side cleanup complete; this PR is now ready for review.

Verification:

• GitHub checks are green: test-core, test-cli, swiftlint, shellcheck, update_release_draft, and pr-clean-gate.
• scripts/ci/check-pr-clean.sh osaurus-ai/osaurus 988 reports clean.
• Local verification passed for both Packages/OsaurusCore and the standalone Packages/OsaurusEvals build.

The previous failing run was the stale DerivedData/EventSource dependency class, not the eval smoke-suite changes. Current head includes the same CI hardening and FluidAudio 0.14.3 lock alignment as the other green recovery branches.