Add native LLM request adapter

[kitlangton]

Summary

• add an offline adapter that converts normalized session LLM inputs into native @opencode-ai/llm requests
• map provider package metadata to native route ids, model refs, messages, tools, generation options, provider options, and headers
• cover the adapter with focused request-shape tests without changing the live AI SDK stream path

Test Plan

• bun test --timeout 5000 test/session/llm-native.test.ts
• bun run test -- test/session/llm-native.test.ts
• bun typecheck
• bunx prettier --write src/session/llm-native.ts test/session/llm-native.test.ts
• bunx oxlint packages/opencode/src/session/llm-native.ts packages/opencode/test/session/llm-native.test.ts from repo root
• git diff --check

Stack

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