Triton -> CuteDSL dim0 quantization kernel with RCEIL scaling that writes scales directly to ((32,4),4) layout for tcgen05 mma

[danielvegamyhre]

For shape (128000, 7168)

• Current: ~5.5 tb/s
• Goal: ~6.4 tb/s.

Currently writes scales in row major, requires additional lightweight kernel for per group blocked layout. We should write directly to blocked layout.

Also, we should try CuteDSL instead of Triton for this next iteration

[danielvegamyhre]

Added in #4156 which writes scales directly to tcgen05.mma layout, reaching 6.4TB/s for large inputs and overall 1.08x-1.26x faster than the 2 kernel "triton quant + triton scale transform" approach.