Add ELF competitor benchmark iteration direction report

[yvette-carlisle]

Summary

Adds a durable benchmark interpretation report that translates the current competitor-strength matrix into ELF iteration direction without claiming unproven wins.

The new report covers:

• current measured data for fixture-backed ELF aggregate, ELF/qmd live real-world sweep, production-provider/backfill/restore evidence, and external adapter ledger counts
• scenario-level conclusions for retrieval/debug, work resume, project decisions, source-of-truth, temporal memory, consolidation, knowledge pages, operator debugging, capture/write policy, production ops, personalization, context trajectory, core-vs-archival memory, and graph/RAG navigation
• project guidance for all tracked memory/RAG/graph projects in the current manifest
• explicit non-claims and recommended next reports

Scope

Report-only change. This does not implement ELF optimizations, queue Decodex lanes, or change benchmark semantics.

Verification

• cargo make real-world-memory
• git diff --check
• checked that the report mentions all 17 tracked manifest projects