model: Add Byrne-Embed model implementation

[SeanceTable]

Adds a model implementation for Quazim0t0/Byrne-Embed, a compact 85M sentence-embedding model, per the adding-a-model guide.

• New file: mteb/models/model_implementations/byrne_models.py
• ByrneEmbedModel subclasses AbsEncoder and implements encode(); it snapshot_downloads the model repo (which bundles its self-contained byrne_embedder.py loader + modeling code) and produces 768-dim unit-norm embeddings.
• ModelMeta registered with revision matching the submitted results (b1dd734…).
• Model card: https://huggingface.co/Quazim0t0/Byrne-Embed
• Results were submitted in Add Byrne-Embed (Quazim0t0/Byrne-Embed) MTEB(eng, v2) results results#577 (MTEB(eng, v2), 41/41 tasks, overall 50.79).

Verified the meta validates and registers in MODEL_REGISTRY locally.

[Samoed]

You shoudn't use snapshot download. Update to use trust_remote_code or create a library with code

[SeanceTable]

Thank you, and apologies for the snapshot_download approach — that was the wrong call. I've fixed it: the loader now uses trust_remote_code only.

• The model repo (Quazim0t0/Byrne-Embed) is now a proper custom-code transformers model: the projection head is fused into the weights and modeling_byrne_embed.py is fully self-contained, so it loads with a plain AutoModel.from_pretrained(..., trust_remote_code=True) (and AutoTokenizer likewise).
• ByrneEmbedModel no longer does any snapshot_download/sys.path manipulation — see the updated byrne_models.py.
• Verified end-to-end: clean cache → AutoModel/AutoTokenizer via trust_remote_code → encode produces the expected embeddings.
• The revision is pinned to af905dd and the results PR (Add Byrne-Embed (Quazim0t0/Byrne-Embed) MTEB(eng, v2) results results#577) has been updated to the same revision (embeddings are identical to the prior weights, parity 1.0).

Apologies again for the back-and-forth — my ISP is currently blocking the mteb docs site, so I've been navigating the contributing guide on my phone while working on my computer, which made me miss the intended pattern at first. Please let me know if anything else needs adjusting.