Merge pull request #1 from cneiman/claude/distracted-chandrasekhar

fix: safer NOT IN clauses, --user pip install, generic adapter dependency docs

fix: release readiness — security hardening, bug fixes, and doc corrections

Add MIT LICENSE file

docs: add benchmark results (78.0% LLM-judged) and test suite to README

fix: add appendFileSync import to evaluate.js, update .gitignore for venvs and eval data

feat: add core test suite (schema, search, MCP, reranker, integration)

fix: revert temporal and counting prompts, keep preference-only category-aware prompt

feat: category-aware reader prompts for preference, temporal, and counting questions

fix: correct Zep LongMemEval score — 71.2% (gpt-4o), not 63.8% (gpt-4o-mini)

fix: 7+ weeks → 7 weeks (accurate as of 2026-03-19)

fix: replace symlinks with real files, add missing adapters and context script

merge: reunite remote (extraction) with local (evals)

feat: add Groq provider support + hybrid/temporal/rerank pipeline

feat: wire semantic search, temporal reasoning, and reranking into eval harness