Split sending of RC_CHANNELS and RC_CHANNELS_RAW

[peterbarker]

Before this patch is applied we may never send the second message
because there's not room for it in the buffer and we can't return
failure-to-send (always interpreted as "retry") as we're in a void function.

Further, if you are on a mavlink2 connection we will not send out the
RC_CHANNELS_RAW message, depriving the user of any RC_CHANNELS messages.

This patch does have the drawback of doing more work on a mavlink1
connection - it has to fetch the data twice.  On the other hand, it also
allows the GCS to set the message rates independently for both
RC_CHANNELS and RC_CHANNELS_RAW so one or the other can be squelched.
That could be handy for reducing bandwidth if you know you're not using
more than 8 input channels.

[tridge]

I'd like to reproduce the issue we hit that triggered this work so we can confirm it is fixed

[tridge]

i've reproduced, and this PR does fix it