add contrib/dot benchmarks

[mytsvy]

Add a dot product benchmark that measures both generic and kMultipleOfVector paths side-by-side across common sizes.
Dot::Compute has assumption-dependent fast paths; benchmarking only the generic path hides key perf behavior. This adds an explicit comparison in one run.

Test:

[ RUN      ] DotBenchGroup/DotBench.BenchAllDot/AVX2
Dot same (f32x8):
256 elems: gen  163.04 GB/s (MAD 0.29%) | mult  160.82 GB/s (MAD 0.28%)
1024 elems: gen  209.07 GB/s (MAD 0.22%) | mult  193.11 GB/s (MAD 1.06%)
16384 elems: gen   95.18 GB/s (MAD 0.89%) | mult   95.75 GB/s (MAD 0.97%)
262144 elems: gen   54.48 GB/s (MAD 0.47%) | mult   54.50 GB/s (MAD 0.45%)
1048576 elems: gen   41.13 GB/s (MAD 5.64%) | mult   44.42 GB/s (MAD 1.92%)
[        OK ] DotBenchGroup/DotBench.BenchAllDot/AVX2