Summary

Adds lightweight LLM trace summaries for assistant model runs and includes them in local session exports.

Why

Session exports currently preserve final messages, parts, and token metadata, but they do not show whether an output anomaly happened at the request boundary, AI SDK normalized stream, PawWork processor, or persisted message. This PR adds count-only diagnostics so issue #454 can be diagnosed from the export without raw prompts, raw model output, headers, API keys, or provider chunks.

Related Issue

Closes #454
Closes #214

Human Review Status

Pending. A human should make the final merge decision after reviewing the final diff and verification evidence.

Review Focus

Please focus on the trace contract and privacy boundary: request summary allowlist, normalized stream event counts, final stored part counts, and whether assistant-message retention is the right persistence layer.

Risk Notes

Additive assistant message diagnostics widen the v2 SDK schema and persist one count-only trace per assistant model run. The trace intentionally omits prompts, messages, headers, API keys, raw provider chunks, raw output text, and tool bodies. Generated SDK files changed after bun --cwd packages/sdk/js build. No desktop, packaging, updater, signing, path, shell, or permission behavior is changed.

How To Verify

LLM trace/export tests: 35 passed, 0 failed
Processor integration tests: 15 passed, 0 failed
opencode typecheck: passed
SDK build: passed, v2 SDK generated files updated
SDK typecheck: passed
Diff check: no whitespace errors

Commands run:

bun --cwd packages/opencode test test/session/llm-trace.test.ts test/session/export.test.ts --timeout 30000
bun --cwd packages/opencode test test/session/processor-effect.test.ts --timeout 30000
bun --cwd packages/opencode typecheck
bun --cwd packages/sdk/js build
bun --cwd packages/sdk/js typecheck
git diff --check

Screenshots or Recordings

Not applicable. No visible UI changes.

Checklist

• Human review status is stated above as pending, approved, or not required
• I linked the related issue, or stated why there is no issue
• This PR has type, scope, and priority labels, or I requested maintainer labeling
• I described the review focus and any meaningful risks
• I listed the relevant verification steps and the key result for each
• I did not introduce unrelated refactors, dependencies, generated files, or file changes beyond the stated scope
• I manually checked visible UI or copy changes when needed, with screenshots or recordings
• I considered macOS and Windows impact for desktop, packaging, updater, signing, paths, shell, or permissions changes
• I called out docs, release notes, dependencies, permissions, credentials, deletion behavior, generated content, or local file changes when relevant
• I reviewed the final diff for unrelated changes and suspicious dependency changes
• I am targeting dev, and my PR title and commit messages use Conventional Commits in English

Requested maintainer labeling for type, scope, and priority.