Attention kernel optimization

[jikunshang]

• [Decode] acc/perf improvement: Tune attention perf to align with IPEX attention functions #162
• [Decode] causal acc fix. fix is_causal impact on decode kernel #181
• [Decode] sliding window Add sliding window support for paged decode kernel #168
• [Decode] fp8 kv cache Support FP8 KV cache in paged_decode kernel #166
• [Decode] MLA support Support arbitrary KV cache strides in paged_decode for MLA #165
• [Decode][ChunkPrefill] block size support 16/32 Add block_size 16/32 support for chunk prefill and fix paged decode #171
• [Decode][ChunkPrefill] block size support multiple of 32/64
• [Decode][ChunkPrefill] support query use FP8 [fmha] support fp8 query #153

[jikunshang]

cc @YizhouZ @baodii @xinyu-intel