docs: shift vector search to v0.3.0 and make provider-agnostic

- Moved vector search out of v0.2.0 (blocking the TUI release).
- Re-architected as 'Asynchronous Embedding Gateway' in v0.3.0.
- Supports Ollama, llama.cpp, and OpenAI based on .env configuration.
- Operates via a background worker thread to prevent Merkle GC churn during active text editing.

gtd.org

@@ -121,10 +121,6 @@ Roadmap basis: Evolutionary roadmap from README.org. Working backwards from SOTA
       - Create/update/delete headlines, set properties, manage TODO states
       - Handle =id:= links and internal links
       - Pure Lisp implementation (no Emacs subprocess)
-***** TODO Implement local vector search (Ollama embeddings)
-      - =generate-embeddings= via Ollama API
-      - Add =:vector= slot to org-object
-      - Semantic search with cosine similarity over memory
 ***** TODO Implement tool permission tiers (ask/allow/deny)
       - Per-tool permission plist stored in org-object
       - =generate-tool-belt-prompt= filters denied tools before LLM sees them
@@ -217,6 +213,12 @@ Unified control plane: hooks + cron + routing in one skill. Deep project underst
       - =:project= (scoped to current work)
       - Scope-aware retrieval in memory.lisp
 
+**** TODO 4. Asynchronous Embedding Gateway
+***** TODO Implement provider-agnostic org-skill-embedding-gateway
+      - Support Ollama, llama.cpp, and OpenAI based on .env config
+      - Implement lazy-loading: edits mark nodes as =:vector :pending=
+      - Background worker thread batches pending nodes and updates Merkle tree silently
+
 **** TODO 4. Slash Commands (TUI ergonomics)
 ***** TODO M-x style command palette
 ***** TODO /- prefix for command mode