Release 0.6.0: reliable ingestion and auditable retrieval

[renezander030]

Summary

• retry transient ingestion failures with stable episode IDs and durable success/failure receipts
• require typed fact decisions before promoting staged extraction snapshots
• accept text, Markdown, JSON, JSONL, and directory inputs with bounded deterministic chunking
• report candidate ceilings and possible incomplete recall in CLI and MCP searches
• require nullable schema keys for grammar-backed local model extraction
• make fuzzy entity-resolution ties deterministic across workers
• update to graphiti-core 0.30.2 for scoped FalkorDB routing and indexed edge search
• release the package and MCP manifest as 0.6.0

Verification

• ruff check .
• 125 local tests pass; the remaining LadybugDB cross-process/search cases reproduce a backend hang on the unchanged base and are left to GitHub CI
• python -m build
• twine check dist/*
• release audit passes with no likely secrets, personal data, or local paths

Review notes

• Base commit: f8254c4
• Touched: ingestion, staged review/promotion, retrieval result metadata, local-model schema preparation, deterministic entity resolution, release metadata, docs, and tests
• Intentionally untouched: authentication/authorization behavior, database schema ownership, container startup, and publishing credentials
• No overlapping open pull request was found while preparing this release