Inject native LLM client into session service

[kitlangton]

Summary

• make opencode's LLM service consume an injectable native LLMClient.Service
• keep production defaults by wiring LLMClient.layer with RequestExecutor.defaultLayer in LLM.defaultLayer
• add a native runtime test with an injected request executor that scripts a tool call and verifies the tool receives the native call id

Verification

• cd packages/opencode && bun typecheck
• cd packages/opencode && bun run test -- test/session/llm.test.ts --test-name-pattern "injected native|native runtime"
• pre-push root bun turbo typecheck

Stack

1. Consume native LLM events in session processing #26639
2. Add native LLM request adapter #26941
3. Compile native LLM requests in session tests #26946
4. Add native OpenAI runtime opt-in #26947
5. Inject native LLM client into session service #27065 👈 current

[kitlangton]

Superseded by combined preview #27114.