feat(mcp): add multimodal vision tools via MCP sampling

analyze_floorplan_image and analyze_room_photo defer the vision work
to the host via MCP sampling (server.server.createMessage). Validates
host capability before calling, fetches URL inputs and base64-encodes
them, constrains output to a Zod schema, and returns structured
content. No vision model is bundled.

9 tests, all passing via a mocked sampling-capable client.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

packages/mcp/src/tools/vision/index.ts

@@ -0,0 +1,26 @@
+import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js'
+import type { SceneBridge } from '../../bridge/scene-bridge'
+import { registerAnalyzeFloorplanImage } from './analyze-floorplan-image'
+import { registerAnalyzeRoomPhoto } from './analyze-room-photo'
+
+/**
+ * Register the vision-input tools that defer to the MCP host's sampling
+ * capability. No vision model is bundled in this package — if the host does
+ * not advertise `sampling` support, calling either tool returns
+ * `sampling_unavailable`.
+ */
+export function registerVisionTools(server: McpServer, bridge: SceneBridge): void {
+  registerAnalyzeFloorplanImage(server, bridge)
+  registerAnalyzeRoomPhoto(server, bridge)
+}
+
+export {
+  analyzeFloorplanImageInput,
+  analyzeFloorplanImageOutput,
+  registerAnalyzeFloorplanImage,
+} from './analyze-floorplan-image'
+export {
+  analyzeRoomPhotoInput,
+  analyzeRoomPhotoOutput,
+  registerAnalyzeRoomPhoto,
+} from './analyze-room-photo'