v2.3 Refactor 6/N

What's Changed

• [Refactor][6/N] CUDA Learn Notes refactor Part-6 by @DefTruth in #17
• [Refactor][5/N] CUDA Learn Notes refactor Part-6 by @DefTruth in #18
• [LayerNorm][Half] support fp16x8 packed LayerNorm by @DefTruth in #19
• [Reduce][Half] add HALF2 & BFLOAT2 macro by @DefTruth in #21
• [RMSNorm][Half] support fp16x8 packed RMSNorm by @DefTruth in #22
• [Bugfix][Kernel] fixed some kernel blocks calculate errors by @DefTruth in #23
• [Elementwise][Half] support fp16x8 packed Elementwise by @DefTruth in #24
• [Elementwise][Half] support fp16x8 packed Elementwise by @DefTruth in #25
• [RELU][Half] support fp16x8 RELU kernel by @DefTruth in #26
• [RMSNorm] support f16x8_f32 RMSNorm by @DefTruth in #28
• [RMSNorm][Kernel] Add FLOAT2/HALF2_VARIANCE macro by @DefTruth in #29
• [LayerNorm][Kernel] Add HALF2 SUM/SUB/VAR macro by @DefTruth in #30
• [HGEMM] Add slicked_k&t_8x8_sliced_k_f16x4 by @DefTruth in #31
• [HGEMV][Half] support hgemv k32/k128/f16 by @DefTruth in #32
• [FlashAttention] Refactor flash_attn_1_fwd_f32 kernel by @DefTruth in #33
• Bump up to v2.3 by @DefTruth in #34

Full Changelog: v2.2...v2.3