Add coding-agent memory dataset seed

[koriyoshi2041]

What

• add a small codingagent synthetic dataset with three multi-session engineering-memory cases
• cover CI recurrence, review preference reuse, and stale docs/release boundary behavior
• register the dataset with the existing CLI registry

Why

This gives the coding-agent memory evaluation discussion in vectorize-io/hindsight#2347 a small runnable seed inside the benchmark harness, without changing scoring or runner behavior.

Checks

• uv run omb providers
• uv run omb splits --dataset codingagent
• uv run omb dataset-stats --dataset codingagent
• lightweight loader/BM25 smoke check via uv run python
• uv run python -m compileall src/memory_bench/dataset/codingagent.py src/memory_bench/dataset/__init__.py
• git diff --check

[vercel]

@koriyoshi2041 is attempting to deploy a commit to the Vectorize Team on Vercel.

A member of the Team first needs to authorize it.