MHA kernels not triggered in nanoGPT sample

[yizhuoz004]

nanogpt_mlir.txt

[pranavm-nvidia]

We should also add a test, maybe using the engine inspector, to make sure that MHA kernels are used where we expect them to be. A perf test would also be a good way to do this since we do ultimately only care about the perf and not really which kernel is used.