WAL checkpointing: Explicitly checkpoint after common/large writes

[Qqwy]

Follow-up of #89

• Ensures we use the periodic checkpointing time interval in Skip mode rather than Burst
• Always checkpoint after inserting a new submission. This is a protection against the situations in which submissions with many thousands of chunks are added, which is the failure we saw today.
• Checkpoint every 100 chunks that are completed.

The last two situations already have a write connection at the time they get to running the checkpoint query, which will make checkpoint-starvation (which in all likelihood is what we saw before) improbable.