feat/jina embeddings v4

[Claude-kang]

What does this PR do?

This PR adds full support for Jina Embeddings v4 across routing, backend handling, and documentation.

Summary of changes

• Model detection & routing

  • Detects JinaEmbeddingsV4Model and text_config fields in config.json
  • Applies sliding_window logic to respect max input length
  • Treats Jina v4 as an embedding model with mean pooling by default in the router

• Python backend support

  • Adds a Python backend handler for Jina v4 (remote code + LoRA support)
  • Returns single_vec_emb
  • Supports automatic task selection, with optional override via JINA_V4_TASK

• Build & test improvements

  • Improves default dtype handling for gRPC-only builds
  • Fixes tokenizer test to use the Tokio API

• Documentation

  • Documents Jina v4 usage in the README
  • Adds Jina v4 to the supported models list

Fixes # (no issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Was this discussed/approved via a GitHub issue or the forum? (No)
• Did you make sure to update the documentation with your changes? (README + supported_models)
• Did you write any new necessary tests?
  - Workspace tests
  - Jina integration tests
  - Tokenizer test updated to use Tokio API

Who can review?

@Narsil
@alvarobartt

[michaelfeil]

should not be removed.

[michaelfeil]

why not implement JinaV4Model in candle? Its much faster than torch here, actually (because you have to use PaddedBatch)