fix(runtime): preserve custom OpenAI-compatible model IDs

[jsiu93]

Summary

Fixes #1195.

• Preserve namespaced OpenAI-compatible model IDs when provider = "openai" uses a custom base_url.
• Route request construction, manifest normalization, model switching, and fallback chains through api_model_name.
• Copy configured [provider_urls] into spawned manifests before normalization so custom endpoint semantics survive agent registration.

Root cause

openai/<model> was treated as an internal catalog prefix everywhere. For custom OpenAI-compatible endpoints such as Featherless, the provider expects the full ID, for example openai/gpt-oss-120b.

Testing

• cargo test -p openfang-runtime --lib
• cargo test -p openfang-kernel --lib
• cargo test --workspace
• cargo clippy --workspace --all-targets -- -D warnings

Security

• No new unsafe code
• No secrets or API keys in diff
• No new user input boundary