XY-861: [ELF benchmark vNext P1] Encode project-decision real-world memory suite

[yvette-carlisle]

Summary:

• Add checked-in project_decisions real-world memory fixtures for accepted decisions, reversals, current validation policy, tradeoff rationale, and bounded caveats.
• Surface per-job answer_type, requires_caveat, requires_refusal, and can_answer_unknown in benchmark reports.
• Add Makefile/docs/test coverage for the project_decisions suite and aggregate expectations.

Validation:

• cargo make fmt
• cargo make lint-fix
• cargo make checks
• cargo make real-world-memory

Benchmark readback:

• project_decisions: status=pass, encoded_job_count=5, score_mean=1.0, expected_evidence_recall=1.0