NOISSUE - Add OpenAI embedding dims and OCR preprocessing

[fbugarski]

What type of PR is this?

This is primarily a feature PR (adds OCR preprocessing and runtime configuration), with a small bug fix/improvement (OpenAI embedding dimensions support/validation), and a documentation update.

What does this do?

This PR improves the Embedder ingestion and model-integration flow by adding:

• OpenAI embedding request dimensions support (and response-dimension validation) to keep compatibility with existing vector(768) usage.
• OCR preprocessing pipeline for ingestion:
  • OCR for image/* files using tesseract.
  • OCR fallback for PDFs (pdftoppm + tesseract) when pdftotext output is too small.
• OCR configuration via environment variables (EMBEDDER_OCR_*) and startup preflight checks for OCR binaries.
• Docker/runtime updates for OCR dependencies (tesseract-ocr, tesseract-ocr-eng).
• Compose and .env wiring for OCR-related settings.
• Documentation updates in Embedder README and workflow docs.
• Added/updated tests for OpenAI embedding behavior and OCR extraction behavior.

Behavior notes:

• OCR is enabled by config and defaults are safe/configurable.
• No API contract breaking changes were introduced.

Which issue(s) does this PR fix/relate to?

• Related Issue: NOISSUE

Have you included tests for your changes?

Yes.

Added tests:

• internal/embedder/embedding/openai/client_test.go
• internal/embedder/ingest/ocr_test.go

Executed locally:

• go test ./internal/embedder/embedding/... -v
• go test ./internal/embedder/ingest/... -v
• go test ./cmd/embedder/...

Did you document any new/modified features?

Yes.

Documentation updated:

• internal/embedder/README.md (new OCR env vars and behavior)
• internal/embedder/workflows/ollama-vs-openai-eval.md (evaluation flow and OCR notes)

Notes

• OpenAI API keys are intentionally not committed in repo config.
• OpenAI live API validation may require active API quota/billing on the target project.
• OCR dependencies are now present in docker/Dockerfile.embedder, so enabling OCR in deployment only requires env configuration.