Fix parsing of 'vmov Q<n>.F32,Q<n>.F32' instruction.

parse_qfloat_immediate() accidentaly parses register with size
qualifier as immediate constant (It takes '<n>.' substring as
valid floating point constant).

Due to this, slightly reorder cases in parse_neon_mov() and move parsing of
vmov with immediate constant to last place.

MFC after: 2 weeks