Set Windows timeouts to enforce non-blocking read

[larsch]

This is the equivalent of the PR serialport/serialport-rs#79. I would actually recommend completely removing the COMMTIMEOUTS override in mio-serial when/if the patch is accepted on serialport-rs, as this will solve the issue of read not returning available data until buffer is full or timeout. However, without this patch, the fix in serialport-rs does nothing.

This behavior is critical for and request/response protocol. I see 27-30 millisecond minimum latency on reads for data that takes <1ms to transfer, which is unacceptable.

Original change description:

Settings ReadIntervalTimeout and ReadTotalTimeoutMultiplier to MAXDWORD in the COMMTIMEOUTS structure on Windows makes Windows behave as normally expected; a read call to the serial port will return immediately if there is any data available, or if a single byte arrives in the buffer. If no data arrives, it will timeout out after the timeout specified in ReadTotalTimeoutConstant.

This fixes issues where reading from a serial port would take longer than desired when less data arrives that there is room for in the buffer passed to the read function.

See also: https://learn.microsoft.com/en-us/windows/win32/api/winbase/ns-winbase-commtimeouts

[estokes]

Ping me when this is accepted in serialport and I'll give it a detailed look