chore(aihubmix): sync model catalog

[EriDeLee]

Summary

Data-only update scoped to providers/aihubmix/models.
Adds newly available models, disambiguates duplicated upstream model families with provider-qualified IDs, updates capabilities/pricing/limits/modalities for existing entries, and removes stale or non-exported entries.

Changes

Added AIHubMix models

• Alibaba Cloud variants:
  • alicloud-deepseek-v4-flash
  • alicloud-deepseek-v4-pro
  • alicloud-glm-5.1
• Doubao Seed 2.0 variants:
  • doubao-seed-2-0-code-preview
  • doubao-seed-2-0-pro
  • doubao-seed-2-0-lite-260428
  • doubao-seed-2-0-mini-260428
• Xiaomi MiMo-V2.5 variants:
  • xiaomi-mimo-v2.5
  • xiaomi-mimo-v2.5-pro
  • xiaomi-mimo-v2.5-free
  • xiaomi-mimo-v2.5-pro-free
  • coding-xiaomi-mimo-v2.5
  • coding-xiaomi-mimo-v2.5-pro

Renamed entries to avoid ambiguous provider IDs

AIHubMix serves multiple variants with similar upstream model names. These renames ensure the model ID clearly indicates the routed backend/provider while remaining under the aihubmix provider:

• In particular, deepseek-v4-flash has been split into two clear branches:

graph TD
    subgraph After ["After: Explicit Provider Branches"]
        deep[deep-deepseek-v4-flash] --> DeepSeek
        ali[alicloud-deepseek-v4-flash] --> Alibaba[Alibaba Cloud]
    end

    subgraph Before ["Before: Random Routing"]
        original[deepseek-v4-flash] --> random[Randomly routed to providers]
    end

Loading

• deepseek-v4-pro → deep-deepseek-v4-pro (routed to DeepSeek)
• glm-5.1 → zai-glm-5.1 (routed to Z.ai)

Updated existing metadata

• Updated Claude Opus/Sonnet 4.x entries (structured output, larger context/output limits, PDF modality, >200k pricing)
• Updated GPT-5.x and Codex entries (release dates, knowledge cutoffs, temperature/structured output support, limits, PDF modality, tiered pricing)
• Updated Gemini entries (pricing details, audio input pricing, cache pricing, long-context tiers)
• Updated Kimi K2.5/K2.6 (limits, attachment/temperature flags, pricing normalization)
• Updated MiniMax M2.7 and coding variants (structured output, interleaved reasoning metadata, cache write pricing, corrected output limits)
• Updated Qwen3.6 entries (more specific family, normalized prices, >200k pricing)
• Updated GLM family/open-weights metadata to reflect AIHubMix vendor variants

Removed stale or non-exported entries

• Removed outdated AIHubMix OpenAI entries no longer present in the synced catalog: gpt-4.1, gpt-4.1-mini, o4-mini
• Removed stray non-TOML file kimi-k2-thinking (model files are only loaded from *.toml)