Consume native LLM events in session processing

[kitlangton]

Summary

• Maps AI SDK fullStream events into @opencode-ai/llm LLMEvents at the existing LLM.Service.stream boundary.
• Updates SessionProcessor to consume native LLM lifecycle event names and shapes directly.
• Moves session usage accounting to native Usage and updates compaction fixtures accordingly.

Stack

• Depends on Add explicit LLM stream lifecycle events #26722 for the richer @opencode-ai/llm stream lifecycle event model.

Tests

• cd packages/opencode && bun typecheck
• cd packages/opencode && bun run test test/session/processor-effect.test.ts
• cd packages/opencode && bun run test test/session/prompt.test.ts
• cd packages/opencode && bun run test test/session/message-v2.test.ts test/session/retry.test.ts test/provider/transform.test.ts
• git diff --check

Stack

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

[kitlangton]

Let's get rid of this alias.