Buffer fillup issue with 20+ hz data stream

[mikepurvis]

It appears that the call to select/pselect in Linux yields the thread, even when data is immediately available, and it should be a noop. I can't find documentation of this behaviour, but it's very clearly evident by monitoring serial_->available() on a sensor which returns data sentences at 50Hz.

I believe the fix is to prepopulate the read buffer from a ::read call prior to the first select call. This prevents select from ever being invoked if sufficient data is immediately available.

This issue affects both the master branch as well as the timespec refactor. I've added the fix for this in a9bf8d8, on the timespec refactor branch (ref #45). Would prefer not to bother with back-porting it.

[wjwwood]

The current implementation assumes that select/pselect does not do this, and if it does, then it is not obvious from the documentation.

A quick ::read before calling select/pselect seems like a low risk solution.

[mikepurvis]

Okay, we can close this once #45 is merged.

I'm hoping that it may be possible to demonstrate the necessity of this in a test, but who knows.