XY-933: [ELF benchmark P1] Add live capture and write-policy scoring

[yvette-carlisle]

Summary

• add live ELF capture/write-policy materialization and scoring for redaction, exclusions, source ids, evidence binding, and no secret leakage
• add capture fixtures plus runtime source_ref validation so capture pass/fail depends on live returned metadata
• refresh benchmark reports and claim boundaries for ELF/qmd/agentmemory/claude-mem capture coverage

Validation

• jq empty docs/research/2026-06-11-capture-write-policy-live-report.json apps/elf-eval/fixtures/real_world_external_adapters/memory_projects_manifest.json
• cargo test -p elf-eval --test real_world_job_benchmark capture -- --nocapture
• cargo make real-world-memory
• cargo make real-world-memory-live-adapters
• cargo make fmt
• cargo make lint-fix
• cargo make checks