fix(sdk-python): log wrap record failures instead of raising

[negillett]

Summary

• When wrap() completes successfully but outbox recording fails, return
  the callable result and log [intentproof] execution record failed.
• Preserve existing behavior when the wrapped callable raises: recording
  failures are still chained via raise ... from record_exc.

Test plan

• pytest -q (91 tests, 100% coverage on touched modules)
• Updated sync and async tests for success-path record failures

[cursor]

PR Summary

Medium Risk
Changes runtime error-handling in wrap() by swallowing execution-recording failures on success paths, which could hide telemetry/outbox outages unless logs are monitored. Application exceptions are still preserved and chained when recording also fails.

Overview
wrap() (sync + async) no longer raises when _record_execution fails after the wrapped callable returns successfully; it returns the callable result and emits a warning via intentproof.instrumentation.

When the wrapped callable raises, behavior is preserved: recording failures are still chained as the __cause__. Tests are updated to assert the new log-and-return behavior for success-path record failures.

^{Reviewed by Cursor Bugbot for commit 1a467e4. Bugbot is set up for automated code reviews on this repo. Configure here.}