server : allow RANK pooling batch splitting for causal LLM rerankers (ie. Qwen3 and Qwen3-VL) (#28876)

* server : allow splitting RANK pooling for causal LLM rerankers

Rerank models fall into two categories: bidirectional cross-encoders
(BERT, etc.) that require all tokens in a single physical batch, and
causal LLMs repurposed as rerankers (Qwen3, Qwen3-VL) that can use
chunked prefill like any other decoder.

Previously the server rejected all RANK-pooling inputs larger than
n_ubatch, and the graph builder hardcoded QWEN3/QWEN3VL arch checks to
determine last-token pooling. This broke long-document and multimodal
reranking for causal models.

Fix: expose llama_get_causal_attn(ctx) so the server can check the
effective runtime attention type (reflecting any --attention override
or set_causal_attn call). Also expose llama_model_is_causal(model)
for querying the static architectural property from GGUF metadata.

can_split() now permits chunked prefill for RANK pooling when the
context is causal. The graph builder's inline arch check is replaced
with the same cparams.causal_attn predicate, removing the duplication.

Assisted-by: Opencode/Qwen3.8-27B

* remove unused llama_model_is_causal, fix whitespace

Assisted-by: opencode

---------

Co-authored-by: timothywang21 <timothywang21@users.noreply.github.com>

tools/server/server-context.cpp

@@ -435,15 +435,26 @@ struct server_slot {
         return task->need_embd();
     }
 
-    // if the context does not have a memory module then all embeddings have to be computed within a single ubatch
-    // also we cannot split if the pooling would require any past tokens
-    // (MTP supports splitting — uses task->need_embd() not need_embd())
     bool can_split() const {
         GGML_ASSERT(task);
-
-        return
-            !task->need_embd() ||
-            (llama_get_memory(ctx_tgt) && llama_pooling_type(ctx_tgt) == LLAMA_POOLING_TYPE_LAST);
+        // MTP supports splitting - uses task->need_embd() not need_embd()
+        if (!task->need_embd()) {
+            return true;
+        }
+        // if the context does not have a memory module then all embeddings have to be computed within a single ubatch
+        if (!llama_get_memory(ctx_tgt)) {
+            return false;
+        }
+        // context can be chunked/split if the pooling type is LAST
+        const auto pooling = llama_pooling_type(ctx_tgt);
+        if (pooling == LLAMA_POOLING_TYPE_LAST) {
+            return true;
+        }
+        // causal rerankers read the last token and have a KV cache, so they can also be chunked/split.
+        if (pooling == LLAMA_POOLING_TYPE_RANK && llama_get_causal_attn(ctx_tgt)) {
+            return true;
+        }
+        return false;
     }
 
     bool can_batch_with(server_slot & other_slot) const {