Add v0.2.0 features: vector search + tool permissions

- Local vector search: Ollama embeddings + semantic search
  - get-embedding with caching
  - cosine-similarity computation
  - semantic-search cognitive tool
  - :semantic-search tool for LLM queries

- Tool permission tiers: security gating for cognitive tools
  - Three tiers: :allow, :deny, :ask
  - Gate in execute-tool-action before tool runs
  - Defaults: :deny for shell/delete-file, :ask for eval/write-file
  - :tool-permissions cognitive tool for management

- Embedding provider support: Ollama AND llama.cpp
  - EMBEDDING_PROVIDER env var
  - EMBEDDING_MODEL env var
  - LLAMA_HOST for llama.cpp server

- .env.example: Add embedding config variables
- Fix parse-message in communication.lisp

- Update ASDF: add test files, tool-permissions skill

All 60 tests pass (6 suites x 100%)

.env.example

@@ -25,7 +25,13 @@ PROVIDER_CASCADE="openrouter,openai,anthropic,groq,gemini-api,ollama"
 OLLAMA_HOST="localhost:11434"
 
 # llama.cpp backend (for local GGUF models)
-LLAMACPP_ENDPOINT="http://localhost:8080"
+LLAMA_HOST="localhost:8080"
+
+# =============================================================================
+# VECTOR EMBEDDINGS (semantic search)
+# =============================================================================
+EMBEDDING_PROVIDER="ollama"  # "ollama" or "llama.cpp"
+EMBEDDING_MODEL="nomic-embed-text"  # model name for embeddings
 
 # =============================================================================
 # MESSAGING GATEWAYS (optional)