This work is one result of WT-17617.  Create the following improvements for the dhandle benchmark for WT (bench/dhandle).

Support for disagg (using the right config options, including PALite, timestamps), and a limited form of predictable replay good enough to create checkpoints on a leader and pick them up on a follower.

Plateaus: instead of slowly increasing the number of dhandles - do some measurements at 0 tables created, then stop measuring and as quickly as possible create 10000 tables, measure for a while, then stop, create 10000 more tables, etc.

Other minor improvements for usability - easier to read output.  Ability to specify fractional percent inputs, more reasonable defaults for number of creates.  Percent for drop operations. Use "checkpoint_wait" for drops.

track non-cache memory usage.