[NVPTX] Merge consecutive elements while buffering constant vectors with sub-byte datatype. (#183628)
NVPTXAsmPrinter currently buffers one element at a time while generating
PTX for constant global vectors. This causes an assertion when dealing
with sub-byte datatype (like i4) since we allocate lesser buffer size.
This PR fixes the printer to merge consecutive elements to form a full
byte before buffering.