automate KVS garbage collection

[garlick]

Problem: assuming #4303 is merged, we have a mechanism for offline KVS garbage collection that has to be manually invoked by running flux shutdown --gc. We need a way to trigger this automatically based on an estimate of how productive it would be.

For example, maybe if flux shutdown is invoked without --gc, and the KVS root sequence number is past some threshold, OR the number of jobs purged is beyond some threshold, garbage collection could be automatically selected. Possibly with a prompt e.g.

$ sudo flux shutdown
OK to perform dump/restore for offline KVS garbage collection (Y/N)?

[garlick]

Couple of notes:

• once kvs: add root sequence number to checkpoint object #4446 is resolved, the sequence number counting from 0 to some threshold may be used as an indicator of how "busy" the KVS has been since the last dump.
• the root sequence number resets to zero across GC because the checkpoint is not part of the dump
• The flux restore --checkpoint called from rc1 ensures that a new checkpoint (seq=0) is written during startup after the dump is restored. The kvs module on rank 0 initializes from this when is loaded.
• The sequence threshold should be configurable via TOML [kvs] table and be disabled by default
• The actual GC should be triggered by flux shutdown, to be carried out in rc3, as it is now when --gc option is specified.
• If threshold is exceeded, add a prompt to flux shutdown like: "flux-shutdown: gc threshold exceeded, do you want to perform garbage collection (Y/n)?"
• It is important that GC be initiated by the shutdown command, so that the dump is not inadvertently performed under the systemd stop timeout.i