[Feature] Speculative decoding support lookahead (#9873)

Co-authored-by: a4zhangfei <a4zhangfei@qq.com>
Co-authored-by: Qiaolin-Yu <liin1211@outlook.com>

sgl-kernel/include/sgl_kernel_ops.h

@@ -457,6 +457,16 @@ void verify_tree_greedy(
     at::Tensor target_predict,
     int64_t cuda_stream = 0);
 
+void reconstruct_indices_from_tree_mask(
+    at::Tensor tree_mask,
+    at::Tensor verified_seq_len,
+    at::Tensor positions,             // mutable
+    at::Tensor retrive_index,         // mutable
+    at::Tensor retrive_next_token,    // mutable
+    at::Tensor retrive_next_sibling,  // mutable
+    int64_t batch_size,
+    int64_t draft_token_num);
+
 void build_tree_kernel_efficient(
     at::Tensor parent_list,
     at::Tensor selected_index,