feat(tests): add Mooncake production trace benchmark

[GentleCold]

Summary

• Replay the public Mooncake FAST'25 Tool/Agent trace with released arrival times, token lengths, and prefix-block relationships.
• Support deterministic token prompts, context filtering, source windows, and online latency accounting.
• Add focused schema, timing, prefix-sharing, coverage, and HTTP send-timing tests.

Motivation

DaseR needs a reproducible production-derived online workload rather than only synthetic request streams. The public Mooncake trace provides arrival timing and prefix sharing without exposing prompt text.

Test plan

• Unit tests added / updated
• PYTHONHASHSEED=0 pytest -q -m "not integration" --ignore=tests/integration --ignore=tests/connector/test_daser_connector.py --ignore=tests/connector/test_gds_transfer.py tests/ passes (362 passed)
• Integration tests marked N/A; this PR adds an external service benchmark
• Benchmark smoke: 100 selected rows, 98 eligible/completed, 2 explicit context skips, 0 failures
• pre-commit run --all-files passes

Checklist

• pre-commit run --all-files passes
• Type hints and docstrings on all new/modified functions
• New behavior includes tests
• Existing CPU test suite passes