[FRONTEND] Backport new runtime from master (#706)

This PR merges the new runtime back into the `triton-mlir` branch. This
adds caching and just-in-time compilation functionality to the
triton-mlir project, and paves the way for re-using tests from the
master branch.

python/examples/empty.py

@@ -1,3 +1,5 @@
+import torch
+
 import triton
 import triton.language as tl
 
@@ -7,4 +9,5 @@ def kernel(X, stride_xm, stride_xn, BLOCK: tl.constexpr):
     pass
 
 
-ret = triton.compile(kernel, "*fp32,i32,i32", constants={"BLOCK": 256}, output="ttgir")
+X = torch.randn(1, device="cuda")
+pgm = kernel[(1,)](X, 1, 1, BLOCK=1024)