Windows, crash handler : better analysis for deadlocks.

in some situations, e.g "Long semaphore wait" crashes by innodb (e.g MDEV-15631), the information provided by the current crash handler is not sufficient. On a deadlock, It is better to write either all threads callstacks.

The easiest way to fix would be to change he default for on Windows to core-file=1, because by default, windows minidumps are really tiny, and only contain stacks and info about loaded modules.

Or we could force minidump writing on Windows , by changing value of core_file from inside Innodb, whenever it detects a deadlock.

Or we could write all stacks trace in this situation, which is almost a no-brainer on Windows (but a brainer elsewhere