Stabilize CP shared-KV prefetch around draft cache hits

Cache-hit EAGLE/NextN draft extends can enter the draft DeepEP MoE immediately after CP shared-KV attention. The partial current-reuse path is kept for target layers, but draft cache-hit suffixes now use full materialization until draft has an explicit same-layer reuse contract. Next-layer MLA/index prefetch is also gated by the actual model depth, so the single-layer draft model does not enqueue unused next-layer async work.

The temporary stage traces used to isolate the hang are removed. The retained draft current-reuse fallback is a bounded warning because it changes the runtime path intentionally.

Constraint: EAGLE/NextN has one executable draft layer and mirrors target KV state.

Rejected: Keep partial current reuse for draft cache-hit suffixes | reproduced hangs at draft layer0 before DeepEP MoE completion.

Rejected: Keep temporary stage traces | useful for diagnosis but too noisy for normal runs.

Confidence: medium

Scope-risk: moderate

Directive: Do not re-enable draft cache-hit partial current reuse without an explicit draft same-layer reuse contract and ETE validation with CP shared KV + HiCache + EAGLE.

Tested: py_compile on edited Python files; git diff --check; temp trace grep returned no matches.

Not-tested: Local targeted pytest is blocked by missing pybase64 in this environment; full ETE after log cleanup not run.

python/sglang/srt/models/deepseek_v2.py

@@ -1680,14 +1680,28 @@ class DeepseekV2DecoderLayer(nn.Module):
             quant_format,
         )
 
-        hidden_states = self.self_attn(
-            positions=positions,
-            hidden_states=hidden_states,
-            forward_batch=forward_batch,
-            zero_allocator=zero_allocator,
-            llama_4_scaling=llama_4_scaling,
-            layer_scatter_modes=self.layer_scatter_modes,
+        previous_cp_shared_kv_num_model_layers = getattr(
+            forward_batch, "cp_shared_kv_num_model_layers", None
         )
+        forward_batch.cp_shared_kv_num_model_layers = (
+            1 if self.is_nextn else self.config.num_hidden_layers
+        )
+        try:
+            hidden_states = self.self_attn(
+                positions=positions,
+                hidden_states=hidden_states,
+                forward_batch=forward_batch,
+                zero_allocator=zero_allocator,
+                llama_4_scaling=llama_4_scaling,
+                layer_scatter_modes=self.layer_scatter_modes,
+            )
+        finally:
+            if previous_cp_shared_kv_num_model_layers is None:
+                delattr(forward_batch, "cp_shared_kv_num_model_layers")
+            else:
+                forward_batch.cp_shared_kv_num_model_layers = (
+                    previous_cp_shared_kv_num_model_layers
+                )
 
         hidden_states, residual = self.layer_communicator.prepare_mlp(
             hidden_states, residual, forward_batch