Gemm Universal unit tests for MainloopIntelW8A8 API

[rishi-yadav]

Data Type Combination	Tests	Status	Avg Time (ms)	Notes
BF16 → BF16 (BF16 acc)	4	All Pass	1,447	Fast, good precision
FP16 → FP16 (FP16 acc)	4	All Pass	15,162	Slower performance
BF16 → BF16 (FP32 acc)	4	All Pass	729	Best BF16 performance
BF16 → FP32 (FP32 acc)	4	All Pass	710	Mixed precision
FP16 → FP16 (FP32 acc)	4	All Pass	885	Good FP16 performance
FP16 → FP32 (FP32 acc)	4	All Pass	703	Fast mixed precision
S8 → S32	4	All Pass	815	Integer ops
TF32 → FP32	1	Pass	2,029	Single test
FP8 → FP32	4	All Pass	2,036	Emerging format
FP16 + S8 → FP32	4	All Pass	864	Mixed input types
Universal GEMM	3	All Pass	728	Advanced features
FP8 Model Tests	10	All Pass	738	ML model scenarios