[Core] Saturate scalar int-to-8-bit conversions

[XFDG]

Summary

• add saturating scalar int to int8_t and uint8_t NumericConverter specializations
• make width-1 and remainder conversions consistent with the existing packed cvt.pack.sat paths
• cover signed and unsigned values on both sides of their representable ranges

Testing

• B200 (SM100a), CUDA 13.1: current main produced -56 vs 127 for signed N=1/N=2 and 255 vs 0 for unsigned N=1/N=2
• B200 (SM100a), CUDA 13.1: fixed targeted probe builds and runs with exit code 0; N=1 and N=2 both saturate consistently
• cutlass_test_unit_core --gtest_filter=NumericConversion.s32_to_8bit_saturates on B200: 1/1 passed

Contribution disclosure

AI assistance was used for repository research, implementation, and running the checks above.

Fixes #3489