fix(agent): address code review issues

- Add empty goal guard in parser (returns done instead of passthrough)
- Replace `as any` casts in pipeline.ts with proper ActionDecision types
- Add runtime type guards for untrusted LLM output in classifier
- Add intent action to dynamic prompt so UI agent can fire intents

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

server/src/agent/parser.ts

@@ -339,6 +339,7 @@ export function parseGoal(
   caps: DeviceCapabilities
 ): PipelineResult {
   const trimmed = goal.trim();
+  if (!trimmed) return { stage: "parser", type: "done", reason: "Empty goal" };
   for (const matcher of PATTERNS) {
     const result = matcher(trimmed, caps);
     if (result) return result;