Add NanoGPT model sync provider - #2891

nice work!

Conflicts are fixed and PR updated

hopefully this is merged soon!

AUTOMATED REVIEW:

The sync implementation has a promising foundation, but I found several issues that should be addressed before merging:

• inferReasoningOptions() derives controls from model IDs and a boolean capability. The catalog does not describe request controls, and a full sync generates 76 unverified toggles. kimi-k2.7-code even says thinking is forced while declaring a toggle.
• Fifteen added TOMLs duplicate canonical model metadata instead of using base_model, including the TEE Qwen/GLM/DeepSeek, Kimi K2.7, Nemotron, Fugu, MiMo, and GLM 5.2 variants.
• limit.input is preserved while limit.context is refreshed. A current sync produces invalid combinations such as Grok with a 500K context but a 1M input limit, and QwQ with a 32K context but a 128K input limit.
• deleteMissing = false retains removed IDs indefinitely. The live catalog currently lacks 57 local entries, including the newly added xiaomi/mimo-v2.5-pro-ultraspeed.
• Missing pricing values fall back to zero, and nullable cache prices are multiplied into zero, which can make models appear free.
• North Mini Code and Nex N2 Pro are published open-weight models but are marked open_weights = false.
• Current GLM 5.2 pricing is 0.42 input, 1.32 output, and 0.078 cache read, rather than the committed 0.90, 2.70, and 0.20.

bun validate and git diff --check pass, but a normal sync currently reports 25 creations, 547 updates, and 57 retained missing models. The PR only validates --new-only, which hides most of that impact. GitHub also currently marks the branch as conflicting with dev.

The code is worth continuing, but I recommend splitting the sync implementation from the generated catalog migration and adding focused translator tests before reviewing the full-sync output.

Closing this pull request as stale because it has not been updated since feedback from @rekram1-node was left 7 days ago. Feel free to reopen it or submit a new pull request if the work is resumed.