Drastically reduce CPU usage and allocations

[canton7]

SerialPort.Read takes a count parameter, which specifies the maximum
number of bytes to read from the serial port. If this is non-zero, then
the Read call will block until at least 1 byte has been received.

Previously, if there were no bytes available to read, then the code span
in a polling loop until a byte became available, burning CPU. It's more
efficient to call Read with a non-zero count parameter in this case,
which causes the thread to sleep until one or more bytes are available.

Change the Run method to follow the normal pattern of allocating a single
buffer, and then reading some number of bytes into it, and processing
just the bytes which were read. This also avoids the thousands of
allocations of single-byte arrays which were happening previously.

The _sleepTime field is now a bit redundent: I assume this was added
in an attempt to reduce CPU usage. I've left it in out of backwards
compatibility concerns, but it's not adding anything useful any more.

[andywcoder]

Thanks for your contribution and sorry about my late response. Somehow I missed that pull request. Your change makes sense and I will accept the pull request. Please allow me some time until I publish an updated nuget package.