Added Squat Packing's rotate and sum operation and resolved type mismatch issues.

[copybara-service]

Added Squat Packing's rotate and sum operation and resolved type mismatch issues.

[asraa]

Hey, this is a rebase of #970 with a major modification that changes how this pass will be applied later:
I modified the pattern converter to only apply the function signature type conversion / func::ReturnOp type conversion when there was a linalg matmul operation in the body that needed to be squat packed.

• This prevents failure for tests that apply the linalg-to-tensor-ext pass without any matmul operations since the pass promotes the arg shapes
• In the future, I expect @j2kun 's work on layout inference to actually insert operations that change the function args tiling/packing instead of what this pass does - which is just promote the types to "work" (which is OK for the simple examples).

cc @AlexanderViand-Intel

[AlexanderViand-Intel]

Thanks @asraa for rebasing and updating this! I agree that the current solution is more than good enough until #1264 is ready!