Add an API embedder backend (OpenAI / Cohere) behind an extra

[AshwinUgale]

The problem / use case

Context: Embedders live in retrieval_lab.embedding — DeterministicEmbedder
(keyless core) plus E5/BGE behind the [real-embed] extra, imported lazily. An API
embedder is a natural addition for teams comparing hosted vs local models.

What: Add an api embedder (start with one provider) behind a new extra
(e.g. [api-embed]), selectable via --embed-models api:text-embedding-3-small.

Acceptance:

• lazily imported so the core stays dependency-free and the demo stays keyless
• embeddings are cached by content hash (reuse the existing EmbeddingCache)
• needs an API key → skipped/mocked in tests; documented in the install-options + limits
• CHANGELOG.md updated

Note: needs a key to run for real; bigger than a one-file change — hence help wanted.

Proposed idea

Add an API-based embedder (e.g. OpenAI text-embedding-3-*) behind a new optional extra, lazily imported like the E5/BGE path, so users can benchmark hosted embeddings against local ones — without adding a runtime dependency to the core.