Allow job resubmission from within remotely run tasks

[riga]

Adds a way for a remote job to ask to be resubmitted without counting as failed. For example, a task can do this

• when it finds it landed on an unsuitable node, or
• to gracefully shutdown a training once the walltime is close being reaches.

Example:

def run(self):

    ...

    if some_walltime_condition:
        self.request_job_resubmission(
            reason="graceful termination",
            info={"host": socket.gethostname()},
            abort=True,
        )

    ...

• Task.request_job_resubmission() (or law.job.base.request_job_resubmission()) writes the request to $LAW_JOB_RESUBMIT_FILE.
• law_job.sh checks for this file after the setup and after each task branch. If it exists, the job stops and exits with code 100.
• Remote workflows always resubmit these jobs without counting them towards --retries, and list them in their own resubmission summary.