firestore-vector-search: multimodal embedding provider is non-functional in both extension and kit

`EMBEDDING_PROVIDER=multimodal` is offered by both the extension and the kit, but neither produces working embeddings:

- **Extension** ([firestore-vector-search](https://github.com/GoogleCloudPlatform/firebase-extensions/tree/main/firestore-vector-search)): [`MultimodalEmbeddingClient`](https://github.com/GoogleCloudPlatform/firebase-extensions/blob/main/firestore-vector-search/functions/src/embeddings/client/multimodal/index.ts) targets `locations/us-central-1` (invalid region), hardcodes `image/jpeg`, and has no response error handling.
- **Kit** (`kits/firestore-vector-search`): `MultimodalEmbedClient` throws `Multimodal embeddings are not implemented in this package`.

Selecting it deploys an instance that fails on every embedding.

**Options:** implement it properly in both, or remove the value from both. Removing is a breaking config change, so it is deferred until the parity work is done (parity ledger #2974, firestore-vector-search §3).

Related: #3014, #3074 (closed, not pursued for now).

Start by reading the extension's functions/src/embeddings/client/multimodal/index.ts and the kit's MultimodalEmbedClient, then consult parity ledger #2974 §3 and related issues. Trace the failure in both packages and determine the agreed parity path. Done means multimodal selection no longer deploys an instance that fails every embedding, consistently across the extension and kit.

Written by the indexing model from the issue text.