chore: snapshot working tree - pty_exited notifications + in-flight inference WIP

feat(booterm): structured pty_exited WS notifications. Plan-validated, impl-validated, code-reviewed green (contracts build clean, contracts test 29/29, booterm + web typecheck clean).

wip: in-progress inference/provider refactor (agents.ts, provider.ts, new llama-providers.ts, removed llama-args-validator), plus arena, dispatcher, compaction, schema changes.

openspec: pty-exit-notifications complete; x-agent-flags planned (not yet implemented).

apps/control/data/suite-utility-calls.yaml

@@ -0,0 +1,57 @@
+id: utility-calls
+name: Utility Calls
+kind: chat
+version: 1
+description: Titles, summaries, compaction -- directly tunes the FAST_MODEL choice.
+judge_model: null
+tasks:
+  - id: auto-title
+    prompt: "Generate a concise title (max 5 words) for this chat session. The conversation is about: A user asking how to fix a PostgreSQL connection pool exhaustion error in their Express.js application."
+    rubric:
+      criteria:
+        - criterion: relevance
+          description: "Title relates to PostgreSQL connection pool issue"
+          weight: 2
+        - criterion: conciseness
+          description: "5 words or fewer"
+          weight: 2
+        - criterion: clarity
+          description: "Title is specific, not generic"
+          weight: 1
+      max_score: 5
+  - id: chat-summary
+    prompt: "Summarize this conversation in 2-3 sentences: User asked about Docker networking. Assistant explained bridge vs host mode. User asked about port mapping. Assistant showed docker run -p syntax. User confirmed it works."
+    rubric:
+      criteria:
+        - criterion: accuracy
+          description: "Summary captures all key topics discussed"
+          weight: 2
+        - criterion: length
+          description: "2-3 sentences as requested"
+          weight: 1
+        - criterion: readability
+          description: "Flows naturally, not a list of facts"
+          weight: 1
+      max_score: 4
+  - id: context-compaction
+    prompt: "Compress this conversation history into a single paragraph that preserves the essential context for continuing the discussion."
+    rubric:
+      criteria:
+        - criterion: preservation
+          description: "Retains key technical concepts: retry, backoff, circuit breaker"
+          weight: 2
+        - criterion: brevity
+          description: "Single paragraph, significantly shorter than original"
+          weight: 2
+        - criterion: usability
+          description: "Useful context for continuing the conversation"
+          weight: 1
+      max_score: 5
+  - id: label-generation
+    prompt: "Classify this user message into one of these labels: [question, bug-report, feature-request, small-talk, code-review]. Message: 'The app crashes when I click the submit button on the settings page. I'm using Chrome 120 on macOS.'"
+    rubric:
+      criteria:
+        - criterion: accuracy
+          description: "Classifies as 'bug-report'"
+          weight: 3
+      max_score: 3