fix CMS for 224x128x64 NT bf16

[emezh]

Fixes issue #4422 by reducing the dscnt (follow-up from #4137)

No impact to performance in Tensile.

Tested with tensile-client

        - ProblemSizes:
          - Exact: [3584, 2048, 1, 8192]
          - Range: [[224], [128], [1], [1, 16, 64]]
          - Range: [[224], [128], [1], [32, 64, 256]]

hipblaslt-test also passes

[==========] 21891 tests from 12 test suites ran. (1441926 ms total)
[  PASSED  ] 21891 tests.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.