fix: dot/schedule_bench A_stride_m for transpose_a kernels

[kasper0406]

Summary

• schedule_bench was passing a_stride_m (the {i, tile_k} intra-row stride) as the 6th positional arg to kernels that set dot_flag::transpose_a, but for those kernels the 6th arg is consumed as the stride along the k1 dimension of the packed tensor (the advance per inner-k step)
• subgraph/dot.cc::call_kernel already does the correct swap (transposed_a ? a_k_strides[0] : a_stride_m). This change mirrors it in schedule_bench so the bench exercises the same work the production path executes
• The built-in correctness check (A = B = 1, assert c == k everywhere) can't catch this class of bug because the wrong-stride loads still return 1s. Benchmark GFLOPS numbers reported against this bench were inflated by artificial cache hits on overlapping reads

Test plan

• bazel test //ynnpack/kernels/dot/... passes (including schedule_bench_test, schedule_test, consistent_arithmetic_test, test)

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[dsharlet]

Thanks for the fix!