Adds an opt-in 'twelvelabs' GenAI provider implementing the embeddings
role using TwelveLabs' Marengo multimodal model. Text and image inputs
are embedded into a shared vector space, powering semantic search over
event thumbnails.
The provider only implements the embeddings role; descriptions/chat are
left to other providers since Marengo is an embedding model. Marengo
returns 512-dim vectors, which the existing GenAIEmbedding adapter pads
to Frigate's 768-dim search schema.
Includes no-network unit tests and a live test gated on
TWELVELABS_API_KEY.
