folder_change_reader_t should close directory handle before closing thread pool I/O

[avivanoff]

According to CloseThreadpoolIo documentation:

You should close the associated file handle and wait for all outstanding overlapped I/O operations to complete before calling this function. You must not cause any more overlapped I/O operations to occur after calling this function.

Current implementation of folder_change_reader_state::~folder_change_reader_state does the opposite.

[dunhor]

It says "should" not "must," so I'm not immediately convinced there's an observable issue here. The only thing the documentation says you must not do is issue additional I/O operations, which since this is the destructor, should be assumed to be true. It also says to wait for outstanding operations to complete and to call CancelThreadpoolIo, both of which it does.

That said, it's probably not harmful to close the handle, so I'm not opposed to it.