Add Moonshot Kimi For Coding subscription models

[guillaumegay13]

What this changes

Adds the Moonshot "Kimi For Coding" subscription models, filed as #73 (which asked for a way to pick the Kimi coding model, 2.6 vs 2.7).

Following the catalog convention (the file is the model — one YAML per model, selection happens by choosing the model), this adds one -subscription file per Kimi coding model rather than a model selector param:

• kimi-k2.6 — max_completion_tokens, thinking.type (toggleable enabled/disabled), response_format.type
• kimi-k2.7-code — max_completion_tokens, response_format.type (thinking is always on for the k2.7-code variants, so no toggle)
• kimi-k2.7-code-highspeed — same as kimi-k2.7-code

Type of change

• Add a model (new YAML file under models/)

Source

• Kimi/Moonshot chat API: https://platform.moonshot.ai/docs/api/chat (thinking.type toggle is k2.6-only; k2.7-code is reasoning-only; temperature/top_p/penalties are moonshot-v1-only and not exposed by the kimi-k2.x family).
• Coding model ids confirmed from the live GET /v1/models listing.

Smoke-tested live against the real Moonshot API: kimi-k2.6 (thinking enabled + disabled), kimi-k2.7-code (thinking enabled, returns reasoning_content), and kimi-k2.7-code-highspeed all accepted (200).

Closes #73

[vercel]

@guillaumegay13 is attempting to deploy a commit to the Manifest Team on Vercel.

A member of the Team first needs to authorize it.

[guillaumegay13]

Superseded by #80, which bundles the remaining model-param requests into a single PR so the merge cuts one modelparams release instead of four. Same files, same live smoke tests.