Check for WMMA instead of MFMA when assigning datatypes for attention

[dorde-antic]

Motivation

Resolves https://github.com/ROCm/rocMLIR-internal/issues/2142

Technical Details

Checks for WMMA in perfRunner when assigning datatypes for attention.

Test Plan

Weekly CI - Tuning phase

Test Result

CI RUN
Successfully filtered out f32 attention configs on WMMA architecture (check attention tuning results on gfx1100)
Run failed for other issues related to tuning

Submission Checklist

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

[umangyadav]

@dorde-antic CI has failed on Navi3x. It is still picking f32 attentions

[dorde-antic]

@dorde-antic CI has failed on Navi3x. It is still picking f32 attentions

Weird since on Navi4x it didn't (and the logic we use is the same)
It might be related to how we obtain the chip name on navi3x maybe... i'll investigate

[dhernandez0]

@dorde-antic CI has failed on Navi3x. It is still picking f32 attentions

Weird since on Navi4x it didn't (and the logic we use is the same) It might be related to how we obtain the chip name on navi3x maybe... i'll investigate

@dorde-antic This is blocking weekly CI for upstream merge. Please give priority to this PR

[dorde-antic]

@umangyadav I would rather focus on merging this #2123 as a temp solution than merging this PR. #2123 solves both f32 attn thing and thing that we don't try every possible combination on MITuna