Keyboard report should be checked if its transfer finishs successfully.
Otherwise key stroke can be missing when other key event occurs
before the last report transfer is done.

Boot protocol 10ms interval probably causes this problem in case
it receives key events in a row within the period. NKRO protocol
suffers less or nothing due to its interval 1ms.

Should really not declare variables in .h files - since it's included
in different .c files, a proper linker then complains that the same
variable is declared more than once (once for each .c file that the
offending .h is included in).

With matrix_power_down() in suspend event HHKB JP doesn't start up for some reason.
It is unneeded in actual and removed.

Old console sent unneeded empty data every one milli sencond.
After this fix console flushes endpoint data bank every 50ms only when needed.

This reverts commit c17b8a599e.

The commit made compile error and its intention is unclear.