Changing from void return to int return should not change the ABI.

This got lost when '-a' was removed.

- flush command no longer ignored when busy
- commit_len_fifo write pointer bug fixed

fixed ebm manifest and other things screwing with hdlmake. Beware, this change depends on general cores proposed master

USB devices on active open flush any pending data
However, this breaks PCIe passive notification (EB header is dropped)

The unnecessary dependency of the input synchronizers on the FSM
state made it hard for quartus to meet timing requirements.