#3 Harden audit logging and output postcheck to resolve unresolved PR #2 review comments

[Anonymous]

Originally created by: Copilot

This PR addresses the remaining review-thread feedback on app/main.py, app/policy/gate.py, and app/observability/audit.py. The changes remove sensitive-data exposure paths in audit logs and align policy/check behavior with actual runtime behavior.

• Input logging hardening
• handle_input() now runs precheck before run_started logging.

• Replaced raw input logging with safe metadata (classification, input_length).

• Output logging hardening

• run_completed no longer logs full output payload.

• Completion audit event now emits only safe metadata (output_length).

• Postcheck policy enforcement

• postcheck() now performs a minimal sensitive-content check and redacts sensitive output as "[REDACTED]".

• Updated function intent to match behavior.

• Audit logging correctness

• Removed import-time root logger configuration from audit.py (no global logging side effects on import).

• Switched audit event emission from Python dict repr to JSON serialization for consistent machine parsing.

  :::python
  result = precheck_with_warning(user_input)
  log_event(run_id, "run_started", {
  "classification": result["classification"],
  "input_length": len(str(result["input"])),
  })

  checked_output = postcheck(output)
  log_event(run_id, "run_completed", {"output_length": len(str(checked_output))})

[Anonymous]

• assigned_to: MoneyMan421

[Anonymous]

• assigned_to: Copilot