feat(backend): scaffold PodMan agent pipeline + token server

Express server with /health and a LiveKit token endpoint. Module skeletons
for the full loop: vision (Gemini), GitHub fusion, collision detector (pure,
testable), intervention engine, and continual-learning memory store. Approve
native build scripts (esbuild/genai/protobufjs) at the workspace root.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/src/vision/gemini.ts

@@ -0,0 +1,20 @@
+import type { EngineerContext } from '@podman/shared';
+
+/**
+ * Turn a sampled screen frame into a structured EngineerContext using Gemini
+ * vision. This is the headline capability: it produces the pre-push signal
+ * (which file/symbol an engineer is editing) that GitHub cannot see.
+ *
+ * TODO(vision): wire @google/genai, downscale frames, sample ~1fps/on-change.
+ */
+export async function frameToContext(
+  _frame: Uint8Array,
+  meta: { engineerId: string; podId: string },
+): Promise<EngineerContext> {
+  return {
+    engineerId: meta.engineerId,
+    podId: meta.podId,
+    confidence: 0,
+    observedAt: new Date().toISOString(),
+  };
+}