Remove agent evaluation concurrency limits and skip same-day re-evaluation

[xrendan]

Summary

Follow-up to #131, for AgentEvaluateCommitmentJob:

• Weekly scan silently dropped commitments — the job had enqueue_limit: 550, but the scan enqueues ~600. GoodJob drops enqueues past the limit without error, so up to ~50 commitments were never evaluated on Sundays.
• Concurrency control removed entirely — with the enqueue limit gone and perform_limit also dropped, the GoodJob concurrency extension had nothing left to enforce. The include, the good_job_control_concurrency_with block, and the ConcurrencyExceededError retry handler are gone. Concurrency is now bounded by the worker thread count (good_job.max_threads, currently 5).
• Same-day re-evaluation guard — each evaluation is a ~2 minute agent session. The job now skips a commitment whose last_assessed_at is today unless called with force: true, so a re-run of the scan or a duplicate manual enqueue doesn't repeat work.

AgentEvaluateCommitmentJob.perform_now(commitment)               # skips if assessed today
AgentEvaluateCommitmentJob.perform_now(commitment, force: true)  # always runs

Notes

• A full sweep at 5 threads is roughly 4.5 hours. Raising max_threads is the lever to shorten it.
• During a sweep, agent sessions can occupy every worker thread, so feed refreshes and entry jobs queue behind them. A dedicated queue for agent jobs would be the clean fix if that becomes a problem.

Testing

• bin/rubocop — no offenses
• New tests: same-day skip, forced re-run, previous-day run