Benchmark: QMD backend comparison

[bm-clawd]

Part of #10

Add QMD (experimental OpenClaw memory backend) as a comparison provider.

Requirements

• Index the same corpus with QMD (qmd collection add + qmd update + qmd embed)
• Run queries via qmd query --json
• Score with same metrics
• Side-by-side: BM vs builtin vs QMD per category

QMD specifics

• QMD uses BM25 + vectors + reranking (local GGUF models)
• Has query expansion via local LLM
• Fully local, no API calls
• First run downloads models (~1GB)
• Requires bun + sqlite with extensions

Depends on

• Benchmark: build realistic test corpus from agent memory patterns #11 (corpus)
• Benchmark: eval harness for Basic Memory search #12 (eval harness)
• Benchmark: OpenClaw builtin memory-core comparison #13 (builtin comparison, for consistent framework)

Notes

• QMD may not be installed in CI — make this comparison optional
• Document QMD setup requirements clearly