add task_failure_reason to metadata

[mt-ob]

Distinguish the task that failed on its own merit (the "root cause") from the tasks that were simply terminated as collateral damage.

metaflow.task_fail_reason is the new field and can have the following values:

• exception --> exception in the user's step code
• signal_sigint --> CTRL+C
• signal_sigterm --> external termination signal by K8s etc.
• killed --> reported by orchestrator for flagging the tasks it terminates as collateral damage.

Test Cases:

1. One task in a foreach failed due to a Python Exception.

Task: 3 -> Status: False, Failure Reason: exception  <-- Root Cause
Task: 4 -> Status: False, Failure Reason: killed      <-- Collateral Damage
Task: 2 -> Status: False, Failure Reason: killed      <-- Collateral Damage

2. User Interruption (Ctrl+C)
   All tasks that were active at the time of the interruption were correctly marked with signal_sigint, attributing the failure to a direct user action.

3. A task failed with an exception, and immediately after, the run was interrupted with Ctrl+C before the runtime's cleanup completed.

Task: 3 -> Status: False, Failure Reason: exception
Task: 4 -> Status: False, Failure Reason: signal_sigint
Task: 2 -> Status: False, Failure Reason: signal_sigint

[mt-ob]

Supersedes #1731

[mt-ob]

things can be verified using this snippet:

>>> from metaflow import Flow, Run
>>> latest_run = Run("FailureTestFlow/1756199905447021")
>>> for task in latest_run['fail_step']:
...     reason = next((datum.value for datum in task.metadata if datum.type == 'metaflow.task_fail_reason'), None)
...     print(f"Task: {task.id} -> Status: {task.successful}, Failure Reason: {reason}")