Skip stride-alignment check for size==1 dims in verifyBDSizesStrides

A dim with size==1 is never stepped, so its stride is irrelevant to
the hardware encoder. The previous check rejected legal BDs with
padding `(1,1)` dims on sub-word element types (e.g. i16 on a 4-byte
granularity tile): `1 element * 2 bytes = 2 bytes` tripped the
"not divisible by 4" rule even though the dim never advances.

Caught by test/python/dma_op.py once its innermost dim was made
realizable in the previous commit.

Co-Authored-By: Claude Opus 4 (1M context) <noreply@anthropic.com>

Author: hunhoffe

lib/Dialect/AIE/IR/AIEDialect.cpp

@@ -2241,6 +2241,8 @@ mlir::LogicalResult xilinx::AIE::verifyBDSizesStrides(
   for (int i = 0; i < n; ++i) {
     if (i == 0 && inputStrides[i] == 1)
       continue;
+    if (inputSizes[i] == 1)
+      continue;
     if (inputStrides[i] * elemWidthBits % addressGranularityBits != 0) {
       std::stringstream msg;
       msg << "Stride " << i << " is " << inputStrides[i] << " elements * "