#23172: feat: add Jina AI as embedding provider for memory search

## Summary Adds [Jina AI](https://jina.ai) as a first-class embedding provider for memory search, alongside OpenAI, Gemini, and Voyage. **jina-embeddings-v5-text-nano** (239M params) is the top-ranked embedding model under 500M parameters on [MTEB](https://huggingface.co/spaces/mteb/leaderboard): - MTEB English v2: **71.0** avg (vs embeddinggemma-300m: 62.0, voyage-4-nano: 66.4) - MMTEB multilingual: **65.5** avg (vs voyage-4-nano: 58.9) - Half the size of voyage-4-nano (239M vs 480M) with higher scores across retrieval, STS, and reranking <img src="https://jina-ai-gmbh.ghost.io/content/images/2026/02/IMG_1153.JPG" width="800" alt="MMTEB Multilingual Benchmark"> *MMTEB scores vs model size. jina-v5-text models (red) outperform models 2-16x their size. ([source](https://jina.ai/blog/jina-embeddings-v5-text))* <img src="https://jina-ai-gmbh.ghost.io/content/images/2026/02/IMG_1154.JPG" width="800" alt="MTEB English Benchmark"> *MTEB English v2 scores. v5-text-nano (239M) achieves 71.0, matching models with 2x+ parameters. ([source](https://jina.ai/blog/jina-embeddings-v5-text))* | Model | Params | Dim | Max Tokens | MTEB-EN | MMTEB | |-------|--------|-----|------------|---------|-------| | jina-embeddings-v5-text-nano (default) | 239M | 768 | 8192 | 71.0 | 65.5 | | jina-embeddings-v5-text-small | 677M | 1024 | 32768 | 71.7 | 67.0 | Paper: [arXiv:2602.15547](https://arxiv.org/abs/2602.15547) | [Blog](https://jina.ai/blog/jina-embeddings-v5-text) ## Changes - New `embeddings-jina.ts`: provider using Jina's OpenAI-compatible API (`https://api.jina.ai/v1/embeddings`) - Jina uses a `task` parameter (`retrieval.query` / `retrieval.passage`) instead of `input_type` - API key: `JINA_API_KEY` env var or `models.providers.jina.apiKey` - Auto-selection appends jina after voyage (no defaults changed) - 4 unit tests, all passing