[SM6.9] Provide hidden flag to enable elementwise processing of long vectors

[pow2clk]

For debugging purposes as well as to facilitate the creation of testing such as with #7069, it will be useful to have a flag that bypasses native vector generation and enables the previous elementwise and 4-at-a-time loading for HLSL long vectors. When this flag is enabled, elementwise native operators and DXIL intrinsics will only use their scalar variants even when invoked with long vectors. Additionally, the standard versions of rawBufferLoad and rawBufferStore that operate on 4 elements at a time will be used to load the long vectors in those chunks.