Add tests for ck mha fp16 solver

[BrianHarrisonAMD]

Purpose of review:

• Add tests for CK FA V2 FWD inference solver.
• Follow up to Add ck mha fp16 solver #3277

Note:

• Merging these changes into the integration branch to allow for limited diff, but will follow up with a PR for merging the integration branch into develop once complete.
• Testing and solver are limited to FWD inference for now.

[BrianHarrisonAMD]

Note #3302 updates requirement.txt CK hash to the version needed for the integration branch.
Ill wait until #3302 is merged to develop before merging the integration branch to develop in order to align these properly (it'll probably be merged first anyways, but mentioning just in case).

[CAHEK7]

That's only check which does make some sense with the current approach, but the ideal case is specifically prepared data which will allow us to check statistical properties of the dropout - tensor V must have identity matrixes, so it won't affect the data after dropout and we can say that approximately half of the elements are zero for m_bernulliProbability = 0.5.

It's just a side note of how it can be actually checked.

[BrianHarrisonAMD]

Oh interesting idea.
CK also allows us to capture the dropout data.
I was thinking it would make sense to capture, and use that for verification, but I figured we could add that as part of enabling the additional optional params.