Sync NVIDIA endpoint model catalog

[5kahoisaac]

Summary

Sync the NVIDIA provider catalog with NVIDIA Build’s current free and partner endpoint list. This removes deprecated NVIDIA endpoint TOMLs, adds missing sub-provider models, and normalizes NVIDIA slugs to match the live catalog.

Changes

• Added missing NVIDIA endpoint models across sub-providers:
  • Moonshot: kimi-k2.6
  • DeepSeek: deepseek-v4-flash, deepseek-v4-pro, deepseek-v3.2, deepseek-v3.1-terminus
  • Qwen: qwen-image, qwen-image-edit, qwen3.5-122b-a10b
  • Meta, Mistral, Google, OpenAI, BAAI, Upstage, Sarvam, ByteDance, Abacus, Black Forest Labs
  • NVIDIA-native embed, rerank, safety, voice, video, and Nemotron endpoints
• Removed deprecated NVIDIA endpoint entries no longer present in the filtered catalog.
• Fixed NVIDIA slug normalization:
  • llama-3_3-nemotron-super-49b-v1
  • llama-3_3-nemotron-super-49b-v1_5
• Preserved NVIDIA zero-cost endpoint pricing convention.

Verification

• Parsed NVIDIA Build filtered catalog:
  • URL: https://build.nvidia.com/models?filters=nimType%3Anim_type_preview%2CnimType%3Anim_type_upgrade_available&pageSize=96
  • Source: embedded resource data, grouped by publisher
• Verified catalog parity:
  • 92 live NVIDIA entries
  • 92 local NVIDIA TOMLs
  • 0 missing
  • 0 extra
• Verified publisher counts:
  • NVIDIA 31
  • Meta 11
  • Mistral 9
  • Qwen 8
  • Google 6
  • Moonshot 4
  • DeepSeek 4
  • Black Forest Labs 4
  • OpenAI 3
  • Z.ai 2
  • Minimax 2
  • Stepfun, ByteDance, Sarvam, Abacus, Upstage, BAAI: 1 each
• Ran bun validate successfully.

Notes

Remove .serena/.gitignore and .serena/project.yml from this branch before opening the PR unless those local agent config files are intended to be included.

Related Issues

None

Reference

• Nvidia free & partner endpoint list