[mlir] [VectorOps] fixed bug in vector.insert_strided_slice lowering

Summary:
Rationale:
When lowering to LLVM for different rank insert (n vs k), the offset
arrays needs to drop one dimension (becomes n-1), but the strides
array needs to be preserved (remains k). With regression test.
Note that this example was actually in the documentation, so
extra important to do it right :-)

Reviewers: nicolasvasilache, andydavis1, ftynse

Reviewed By: nicolasvasilache, ftynse

Subscribers: Joonsoo, merge_guards_bot, mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, arpith-jacob, mgester, lucyrfox, liufengdb, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D73733

mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVM.cpp

@@ -548,7 +548,7 @@ class VectorInsertStridedSliceOpDifferentRankRewritePattern
     auto stridedSliceInnerOp = rewriter.create<InsertStridedSliceOp>(
         loc, op.source(), extracted,
         getI64SubArray(op.offsets(), /*dropFront=*/rankDiff),
-        getI64SubArray(op.strides(), /*dropFront=*/rankDiff));
+        getI64SubArray(op.strides(), /*dropFront=*/0));
     rewriter.replaceOpWithNewOp<InsertOp>(
         op, stridedSliceInnerOp.getResult(), op.dest(),
         getI64SubArray(op.offsets(), /*dropFront=*/0,