DB history table housekeeping option

[hjoliver]

Currently, the task_events DB table grows without bound.

This is desirable from a provenance perspective, for finite-length science experiments; but probably not desirable for large endlessly-cycling weather prediction workflows (for example).

We should keep all information by default, but should we allow old info to be deleted at some cut-off point behind the active window? (By number of cycle points and/or by cycle point offset and/or by real time offset interval).

[oliver-sanders]

Related tickets:

• database refactor #3872 (would reduce the size of the DB making housekeeping less necessary).
• Provenance data collection #4609 (would require housekeeping to be turned off).