Dump files quickly fill all space if there's a network problem

[adamstallard]

If a node that was down reconnects, it will create dump files for the scorer to process.

If the node is down for more than a few hours, this creates a problem that requires manually deleting dump files to free space.

Instead of requiring the operator to manually delete dump files to recover their node (which isn't documented), the consensus module should be smarter about the number of dumps to create.