Extraction strategies for mnem ingest pipelines.
Provides keyword extraction (KeyBERT-style cooccurrence) and LLM-assisted
extraction traits used by `mnem-ingest`. Trust filters live here too —
each extracted entity carries provenance + a confidence score so the
ingest layer can drop low-trust noise before it reaches the graph.
Part of [mnem](https://github.com/Uranid/mnem) — versioned, mergeable,
content-addressed knowledge graph for AI agent memory.
Apache-2.0.