docs: document OmniRoute OpenAI-compatible provider setup

[diegosouzapw]

Problem

Kimi Code CLI already supports manually configured OpenAI-compatible providers, but the provider
documentation does not show a reproducible gateway configuration for OmniRoute users. The base URL,
model declaration, and environment-variable mapping are easy to misconfigure.

Proposed documentation

• Show OmniRoute through the existing openai_legacy provider, without provider-specific runtime
  code.
• Use http://localhost:20128/v1 as the OpenAI-compatible base URL.
• Declare the model ID and max_context_size manually because custom providers do not perform model
  discovery.
• Explain that openai_legacy reads OPENAI_API_KEY, including how to map an existing
  OMNIROUTE_API_KEY for one process.
• Mention container/remote-host reachability and the existing Responses/Anthropic provider
  alternatives.
• Keep the English and Chinese provider documentation aligned.

Prepared validation

• 47 focused tests passed for LLM creation, OpenAI Chat Completions, and Responses.
• The VitePress documentation build passed.
• git diff --check passed.
• No remote credential or endpoint was used.

The prepared documentation change is under 100 lines and does not alter runtime behavior. If this
direction matches the maintainers' expectations, I can submit the corresponding pull request.