Investigating WAL behavior for supported platforms

[yatli]

Sometimes, a simpe fflush does not guarantee data hitting the disk. The cache on the disk hardware has to be also flushed. Different systems behave different wrt this issue. There's an API for this in NT but to my knowledge, only a global procfs toggle for Linux.

So it means we should at least print warnings in the log, if the toggle is not properly set.

Also, the file buffer should be flushed periodically to make sure that no data lingers for weak WAL.