AtomicMem llm-wiki-compiler adds memory layer

This is less a memory plugin than a compounding knowledge pipeline, and that is the right mental model for research-heavy agents.

• –It shifts Hermes from storing interaction history to curating a domain wiki that can be queried and reused.
• –The `llmwiki ingest` and `query --save` loop turns useful answers into first-class knowledge, which is the real moat.
• –The tradeoff is obvious: source quality, structure, and curation matter more than with simple vector retrieval.
• –Because it is CLI- and provider-agnostic, it fits teams that want local, inspectable artifacts instead of opaque chat memory.
• –It sits in the same conversation as RAG, NotebookLM, and agent memory layers, but with more emphasis on durable synthesis than raw retrieval.