Optimize state cleanup at end-of-window

When using state with a large keyspace, you can end up with a large amount of state cleanup timers set to fire all 1ms after the end of a window.  This can cause a momentary (I've observed 1-3 minute) lag in processing while windmill and the java harness fire and process these cleanup timers.

By spreading the firing over a short period after the end of the window, we can decorrelate the firing of the timers and smooth the load out, resulting in much less impact from state cleanup.

Imported from Jira [BEAM-9308](https://issues.apache.org/jira/browse/BEAM-9308). Original Jira may contain additional context.
Reported by: SteveNiemitz.

Start by reviewing the state cleanup timer scheduling described in the issue, including the windmill and Java harness paths, and consult the linked BEAM-9308 context for additional details. The work is complete when cleanup timers no longer fire in one concentrated burst immediately after the window, reducing the observed processing lag.

Written by the indexing model from the issue text.