feat(chutes): update Kimi K2.5 TEE model capabilities

[alissonlauffer]

Enable reasoning, tool calling, and multimodal input support for the Kimi K2.5 TEE model. Increase context limit from 32k to 262k tokens and output limit from 8k to 65k tokens. Add support for image and video inputs alongside text. Configure interleaved reasoning content field.

[pkarecki]

+1

[pkarecki]

Please merge. Obviously current values are wrong. The worst part is context limit triggering auto-compact every few messages.

[mugnimaestra]

+1

i wonder how it got wrong in the first place, following specs returned from https://llm.chutes.ai/v1/models it should not be that 32k

{
      "id": "moonshotai/Kimi-K2.5-TEE",
      "root": "moonshotai/Kimi-K2.5",
      "price": {
        "input": {
          "tao": 0.00252314883928846,
          "usd": 0.6
        },
        "output": {
          "tao": 0.0126157441964423,
          "usd": 3
        }
      },
      "object": "model",
      "parent": null,
      "created": 1769568145,
      "pricing": {
        "prompt": 0.6,
        "completion": 3
      },
      "chute_id": "2ff25e81-4586-5ec8-b892-3a6f342693d7",
      "owned_by": "vllm",
      "permission": [
        {
          "id": "modelperm-9a5e0f31d83dbb85",
          "group": null,
          "object": "model_permission",
          "created": 1769568145,
          "allow_view": true,
          "is_blocking": false,
          "organization": "*",
          "allow_logprobs": true,
          "allow_sampling": true,
          "allow_fine_tuning": false,
          "allow_create_engine": false,
          "allow_search_indices": false
        }
      ],
      "quantization": "int4",
      "max_model_len": 262144,
      "context_length": 262144,
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "max_output_length": 65535,
      "output_modalities": [
        "text"
      ],
      "supported_features": [
        "json_mode",
        "structured_outputs",
        "tools",
        "reasoning"
      ],
      "confidential_compute": true,
      "supported_sampling_parameters": [
        "temperature",
        "top_p",
        "top_k",
        "repetition_penalty",
        "frequency_penalty",
        "presence_penalty",
        "stop",
        "seed"
      ]
    },

[mugnimaestra]

Great work! I have 3 critical corrections needed for this PR:

1. name: Should be "moonshotai/Kimi-K2.5-TEE" (not "Kimi K2.5 TEE")
2. output limit: Should be 65_535 (not 65,536)
3. structured_output: Should be true (added, not removed/unchanged)

These corrections were included in PR #738 which I just closed to avoid duplication. Please update before merging.

[rekram1-node]

I think name is correct here, you are citing the model id, a name field shouldnt need "/" in it