feat: route model slots to different provider profiles (multi-provider in one session)

[Whiteknight07]

Duplicate Check / 重复检查

• I have searched existing issues and this is not a duplicate / 我已搜索现有 issue，确认非重复

Problem / 问题

Today a profile maps all Claude Code model slots (haiku/sonnet/opus) to a single provider backend (base_url). [profiles.models] only renames the model string — it can't route, e.g., haiku requests to a cheap provider while sonnet/opus go to a premium one.

Use case: I'd like to run one Claude Code session where:

• haiku slot → OpenCode Go (deepseek-v4-flash, API key)
• sonnet/opus slots → ChatGPT Codex subscription (gpt-5.6-terra / gpt-5.6-sol, OAuth)

Proposed Solution / 建议方案

Allow each model slot to optionally reference a different profile:

[[profiles]]
name = "codex-sub"
provider_type = "OpenAIResponses"
base_url = "https://chatgpt.com/backend-api/codex"
default_model = "gpt-5.6-sol"
auth_type = "oauth"
oauth_provider = "openai"

[profiles.slot_profiles]
haiku = "opencode-go"   # route haiku requests to another profile

Implementation sketch (proxy-side, in handle_messages):

1. After resolving the primary profile, inspect body["model"].
2. If it matches a slot that has a slot_profiles override, re-resolve that profile (its adapter, base_url, auth, strip_params).
3. Reuse the existing translation/streaming/failover machinery — only the target ProfileConfig changes.

This keeps the change small: config struct + one routing decision in handler.rs. Launch behavior (ANTHROPIC_DEFAULT_*_MODEL) is unchanged — the model name string is already sent per-slot.

Alternatives Considered / 备选方案

• Separate profiles per provider and switch sessions manually (current workaround — doesn't give mixed slots in one session).
• Smart Router ([router.rules]) — intent-based, but not per-slot and requires a classifier.

Additional Context / 补充信息

Happy to work on this and open a PR with tests + bilingual docs. Let me know if you'd prefer a different config shape.