Add process directive to re-use existing task directory on resume

New feature

Allow automatic resubmission of processes which overrun maximum cluster time but can been checkpointed and therefore resumed.

Usage scenario

Running nextflow on a maintained cluster, maximum job durations may be 72 hours. Many genome assembly jobs will take longer, but the tools (e.g. metaBAT, metaSPADES) will save checkpoints.

Conventional resume behaviour will not help, as all running tasks will be interrupted, and the resume will recommence the failed jobs afresh.

Suggest implementation

Have an option for cluster jobs (e.g. PBSPro) to resubmit within the same working directory if a job did not run to completion.

I am currently trying to do this manually using the feedback loop, but this appears nontrivial for non-singleton inputs, and I am not sure if I can detect a time-overrun job and override the error handling.

No files or tests are named. Start by tracing Nextflow’s resume handling and cluster-job error handling, especially the PBSPro-style executor path mentioned in the issue. Define how a time-overrun, checkpointable task is detected and resubmitted in the same work directory, including non-singleton inputs; done requires an agreed directive and verified behavior.