[mlir][linalg] Fix Linalg runtime verification pass to handle tensors with dimensions of size 0 (#163791)

Runtime verification on Linalg structured ops unconditionally computed
`end - 1` to determine the last iteration index before composing
indexing maps. This caused spurious "negative index" assertion failures
while operating on empty tensors (tensors with a dimension of size 0).

The issue occurs because: 

1. Empty tensors create loop ranges [0, 0) with zero trip count 

2. Computing end - 1 = 0 - 1 = -1 creates a fictitious negative index 

3. The negative index check triggers even though no loop iterations
occur

The fix is to guard all runtime verification with a check that ensures
all loop ranges are non-empty (start < end) before performing any index
arithmetic.

Example MLIR that previously failed:
```mlir
func.func @fill_empty() -> tensor<0xi32> {
  %c0 = arith.constant 0 : i32
  %empty = tensor.empty() : tensor<0xi32>
  %filled = linalg.fill ins(%c0 : i32) outs(%empty : tensor<0xi32>) -> tensor<0xi32>
  return %filled : tensor<0xi32>
}
```

---------

Co-authored-by: Hanumanth Hanumantharayappa <hhanuman@ah-hhanuman-l.dhcp.mathworks.com>

0 files changed, 0 insertions, 0 deletions