[flang] match the actual data size with the KIND (NFC)

[kkwli]

I think the intent is uint8_t instead of uint16_t to match logical(1). Otherwise it needs to swap byte for the big endian environment.

I have a question regarding this test. What will the corresponding Fortran construct look like for MatmulTranspose(result, *yLog, *vLog, ...)? It seems that we are doing

transpose(matrix( 3 by 2 )) * matrix( 1 by 3 )