Grid reshape to increase hit rate

[mmigdal-nv]

Runtimes in ms:
Current tracking-matmul:

SWIZZLE_FACTOR=1 ./bin/nvfuser_tests --gtest_filter=*FusionAmpereMatmul_CUDA*
...
6.80659
5.72541
5.78003
5.336
5.7224
5.59946

With swizzle_factor:

SWIZZLE_FACTOR=4 ./bin/nvfuser_tests --gtest_filter=*FusionAmpereMatmul_CUDA*
...
5.14675
5.06474
5.08704
5.13798
5.12608
5.07923

[mmigdal-nv]

Implemented Here: #87 so closing this one.