feat: instrument evals usage events via openadapt-telemetry

[abrichr]

Summary

• add thin openadapt_evals.telemetry wrapper over openadapt_telemetry.posthog
• instrument benchmark runner with agent_run, agent_run_completed, and action_executed
• instrument oa evals run/mock entrypoints and record_waa_demos.py (demo_recorded)
• add dependency on openadapt-telemetry>=0.2.0

Validation

• ruff check on openadapt_evals/telemetry.py, openadapt_evals/benchmarks/runner.py, openadapt_evals/cli/main.py
• python bytecode compile check for those files + scripts/record_waa_demos.py

Notes

• this PR assumes openadapt-telemetry release with PostHog support is published first