fix: validate embedding model dimensions

[medlouaynjima]

Fixes #9

What changed

• Detect the actual embedding dimension when LocalEmbedder loads a model.
• Expose the loaded dimension through LocalEmbedder.dims.
• Raise a clear error when the model dimension differs from the expected vector(384) schema.
• Add regression tests for matching and mismatched embedding dimensions.
• Document the schema update and re-ingestion requirement when changing EMBED_MODEL.

Validation

• pytest -q tests/test_embeddings.py
• 3 passed
• git diff --check

[github-actions]

Welcome, @medlouaynjima, and thanks for your first pull request to VaultRAG.

A quick look at CONTRIBUTING.md covers how things are set up and run here.

I'll review this shortly. Ask anything in the thread if something is unclear.

[royalpinto007]

The dimension check only fails once because the invalid model and dimension are cached before validation. After the first ValueError, a second dims access returns 768 and embed() can reuse the incompatible model. Please validate before caching, or persist the failure, and add a regression test for repeated access. Everything else looks good.

[royalpinto007]

The repeated dimension validation issue is fixed, but LocalEmbedder._load() no longer returns the validated model, so embed() fails with AttributeError even for a valid 384-dimensional model. Please restore the return value and add a regression test that calls embed().