feat(orchestrator): Phase 2 engine + tokenizer + tool dispatch

Orchestrator engine:
- engine.rs: unified Mistral Conversations API tool loop that emits
  OrchestratorEvent instead of calling Matrix/gRPC directly
- tool_dispatch.rs: ToolSide routing (client vs server tools)
- Memory loading stubbed (migrates in Phase 4)

Server-side tokenizer:
- tokenizer.rs: HuggingFace tokenizers-rs with Mistral's BPE tokenizer
- count_tokens() for accurate usage metrics
- Loads from local tokenizer.json or falls back to bundled vocab
- Config: mistral.tokenizer_path (optional)

No behavior change — engine is wired but not yet called from
sync.rs or session.rs (Phase 2 continuation).

src/config.rs

@@ -102,6 +102,9 @@ pub struct MistralConfig {
     pub research_model: String,
     #[serde(default = "default_max_tool_iterations")]
     pub max_tool_iterations: usize,
+    /// Path to a local `tokenizer.json` file. If unset, downloads from HuggingFace Hub.
+    #[serde(default)]
+    pub tokenizer_path: Option<String>,
 }
 
 #[derive(Debug, Clone, Deserialize)]