feat(embedding): declare a model's width without sending dimensions - #1857

Summary

• OpenAIEmbedder can only override an embedding model's width by also sending that width to the API as the dimensions parameter. That's correct for a Matryoshka model, but it breaks a model of non-default width that isn't Matryoshka-capable: the API rejects the request outright (e.g. nvidia/Nemotron-3-Embed-1B, 2048-wide, served OpenAI-compatible, errors with "does not support Matryoshka embeddings; dimensions must be unset").
• Adds declared_dimensions / REPOWISE_EMBEDDING_DECLARED_DIMS, a second override that sets the declared width without ever sending it. Precedence: dimensions= > REPOWISE_EMBEDDING_DIMS > declared_dimensions= > REPOWISE_EMBEDDING_DECLARED_DIMS > the _DIMS table > 1536. Only the first two are ever sent on the wire.
• The width-mismatch error now names whichever of the three sources actually chose the number, instead of always pointing at _DIMS or REPOWISE_EMBEDDING_DIMS.

Related Issues

None.

Test Plan

• uv run pytest tests/unit/test_persistence/test_openai_embedder.py -q — 52 passed
• uv run pytest tests/unit/test_persistence/ tests/unit/test_providers/ tests/unit/server/mcp/test_embedder_resolution.py tests/unit/cli/test_embedder_key_resolution.py -q — 416 passed
• uv run ruff check . (the actual CI lint gate; ruff format --check isn't run in CI, and this PR's files are clean under it anyway)
• uv run repowise risk main..HEAD:

  Change risk for main..HEAD: touches files that have broken before · 52nd percentile of this repo's fix-bearing files
    feat(embedding): declare a model's width without sending dimensions
    +212 / -36 lines · 2 files · 2 dirs · 2 subsystems · entropy 0.98 · author exp 0
  
  File                                                              Lines   Prior fixes
  packages/core/src/repowise/core/providers/embedding/openai.py       142           1.0
  tests/unit/test_persistence/test_openai_embedder.py                 106           1.0
  
  Diff shape: Typical · 48th percentile of recent commits by size and spread
    About as risky as a typical commit in this repo.
  

  (impacted-tests / health --file need a completed repowise init against this clone; that indexing run did not finish in time and is omitted rather than faked.)

Why this shape, not a _DIMS table entry

The straightforward fix is hardcoding "nemotron-embed-1b": 2048 in _DIMS. That covers exactly one deployment; every other non-Matryoshka model of non-default width hits the identical wall. The existing dimensions= design already treats width as something the operator declares for whatever model they're running, not something the code enumerates by name — declared_dimensions extends that same idea to the one case it didn't cover.

Checklist

• My code follows the project's code style
• I have added tests for new functionality
• All existing tests still pass
• I have updated documentation if needed

Written with AI assistance; measurements and tests were run and verified locally.