Update llvm commit

[yifeizh2]

Update llvm commit to involve the fix for tensor.unpack canonicalization and isaConvolutionOpInterface.

[yifeizh2]

@dchigarev The fixes mentioned in description are essential for the layout propagation PR to pass CI check. The LLVM commit currently selected is chosen with the thought of making minimum modification (which does not lead to any further changes to imex patch files, etc).
On the latest LLVM main branch, features in patch 0007-Move-chunk_size-into-TensorDesc.patch has been merged. Not sure whether it will be better to directly update to the latest LLVM commit.