NetworkTables writes hang if flush is used more often than the update rate

[virtuald]

If a user calls flush lots of times faster than the update rate, and then stops calling flush, the dispatcher won't send new values to the network connections for a potentially long period of time. It does not affect reads.

I ran into this with pynetworktables, but haven't yet verified the issue with ntcore itself. It's pretty easy to see at

ntcore/src/main/native/cpp/Dispatcher.cpp

Line 303 in 3025a18

m_flush_cv.wait_until(flush_lock, timeout_time,

. Every time a flush occurs, the timeout period is incremented by the update rate.

This is a really difficult bug to trigger, and I don't expect many users will run into it, so I don't recommend a fix released until after CMP. For users that may be affected by this bug, a workaround is to continue calling flush periodically.