We should capture the effective degree of parallelism that we get as a result of parallel checkpoints, and set it as a new statistic.

We can express this statistic as checkpoint reconciliation efficiency:

                 time spent in reconciliation
efficiency = ------------------------------------- x 100
               wall-clock time in __bt_sync_file

If we are able to extract a lot of parallelism, the value will be close to the number of checkpoint threads x 100.

This metric will be also useful for single-threaded checkpoints, in which case it will be a value between 0 and 100 that indicates the percentage of __bt_sync_file is spent in reconciliation vs. other tasks, such as the tree traversal.