Add un-cached memory options for matrix A, B, C, D into hipblaslt-bench

[boringmorning]

No description provided.

[boringmorning]

Test on MI300X with command:
./hipblaslt-bench --beta 1 -i 100 -j 10 --algo_method all -m 4096 -n 4096 -k 4096
./hipblaslt-bench --beta 1 -i 100 -j 10 --algo_method all -m 8192 -n 8192 -k 8192

In this test, using un-cached memory does not significantly influence performance.

[boringmorning]

Test on MI300X with command:
./hipblaslt-bench --beta 1 -i 1000 -j 100 --algo_method all -m 128 -n 128 -k 128
./hipblaslt-bench --beta 1 -i 1000 -j 100 --algo_method all -m 512 -n 512 -k 512
./hipblaslt-bench --beta 1 -i 1000 -j 100 --algo_method all -m 1024 -n 1024 -k 1024