Add Docker live baseline benchmark

[yvette-carlisle]

Summary

• Add a Docker-isolated live baseline runner for ELF plus external memory-system baselines.
• Publish a checked-in June 9, 2026 benchmark report and README benchmark snapshot.
• Add a Markdown report publisher so future benchmark JSON runs can be promoted into reviewed docs.

Benchmark Evidence

• ELF provider stress: Qwen3-Embedding-8B, 4096 dimensions, 480 docs, 16 queries, 8/8 checks, pass, 1163s.
• All-project smoke: ELF and qmd pass all encoded checks; agentmemory passes retrieval but fails/incompletes lifecycle checks; mem0, memsearch, and claude-mem miss expected retrieval evidence; OpenViking is incomplete due local embed install failure.

Verification

• cargo make fmt-check
• cargo check -p elf-worker -p elf-eval --bin live_baseline_elf
• cargo test -p elf-providers
• docker compose -f docker-compose.baseline.yml config
• bash -n scripts/live-baseline-benchmark.sh
• bash -n scripts/live-baseline-report-to-md.sh
• ELF_BASELINE_MARKDOWN_REPORT=tmp/live-baseline/rendered-report.md cargo make baseline-live-report
• git diff --check / git diff --cached --check
• semantic drift audit pass: README/docs claims matched generated JSON, Makefile tasks, Docker compose config, and runner/publisher scripts

Notes

The benchmark run artifacts stay under tmp/live-baseline/ and are not committed. The checked-in Markdown report is the reviewed publication artifact.