[Narada Code 1] Add workbench trace materialization verifier spine

[xTRam1]

Summary

Adds the SDK-side M1 verifier spine for Agent Maker Workbench:

• exposes execution_trace_context on CPython and Pyodide Agent.run() responses
• adds shared ExecutionTrace context model support
• adds trace() and Agent.run(trace=True) materialization paths
• adds narada workbench trace materialize, score, and verify
• writes local proof roots with manifest, command ledger, resolved trace files, artifact index, score, and verification reports
• hardens provenance and anti-reward-hacking checks for source-run authority, non-terminal runs, stale roots, missing artifacts, signed URL leaks, secret leaks, static answer maps, self-review markers, and cleanup failures

Validation

• uv run ruff check packages/narada packages/narada-core packages/narada-pyodide → passed
• uv run pytest packages/narada/tests packages/narada-pyodide/tests/test_cloud_browser.py → 85 passed
• Final clean signed-stack proof: /tmp/narada-e2e-runs/20260609T000509Z-stack-m1-clean-trace-proof-20260608170509
• Clean nested APA trace proof: /tmp/narada-e2e-runs/20260609T000658Z-executiontrace-1c09a9402aed4eae9ac6973ea9afd5bb
• Clean SDK request-id materialization proof: /tmp/narada-workbench-runs/m1-clean-sdk-request-materialized-20260609T001137Z
• Independent review subagent verdict: PASS

Codex sibling refs

• frontend: main
• caddie: NaradaAI/caddie#6601
• narada-python-sdk:
• architecture-docs: NaradaAI/architecture-docs#24