This is based on the generic_at parser, with unnecessary stuff
The G1 routinely screws up CRLFs, so the parser needs to account for
that. This parser ignores leading CRLFs (which is what reference-ril
does as well), as well as trailing LFs (which are sometimes left out).
CRs are used as end-of-message indicators. Since we're not bothering
tracking CRLFs, there's also no need for a GARBAGE state, or MULTILINE
This one looks fine to me.