[Intel GPU] Integrate OneDNN SDPA training forward and backward (pytorch#161058)

This PR is the first split PR of pytorch#156272, only contains the OneDNN code. Please help review.

Pending on OneDNN v3.9 commit update. Don't merge.

Pull Request resolved: pytorch#161058
Approved by: https://github.com/guangyey, https://github.com/EikanWang

Author: LuFinch

aten/src/ATen/native/mkldnn/xpu/Attention.cpp

@@ -260,7 +260,7 @@ _scaled_dot_product_fused_attention_overrideable_xpu(
   alloc_with_matching_layout(query, output, output_shape);
   at::Tensor logsumexp, debug_attn_mask; // not supported
 
-  at::native::onednn::gpu_float_sdpa(
+  at::native::onednn::sdpa(
       batch_size,
       seq_len_q,
       seq_len_kv,
@@ -274,7 +274,9 @@ _scaled_dot_product_fused_attention_overrideable_xpu(
       attn_bias,
       is_causal,
       scale.has_value() ? scale.value() : (1.0 / std::sqrt(head_dim_qk)),
-      output);
+      output,
+      false,
+      logsumexp);
 
   // rng not used
   auto philox_seed = at::empty({}, at::dtype(at::kLong));