sve_merge_store_data doesn't account for source vector elements and destination memory elements being different widths

[jj16791]

sve_merge_store_data assumes its to-be-stored data and the memory access are the same width. This is false with one incorrectly implemented example being the st1w {zt.d}, pg, [zn.d{, #imm}] where the source vector elements are of width 64bits whilst the memory elements to be stored are of width 32bits.