Restore flow for longrope ordering

Author: dylanneve1

src/plugins/intel_npu/src/plugin/npuw/llm_infer_request.cpp

@@ -826,8 +826,6 @@ void ov::npuw::LLMInferRequest::infer_prefill(ov::SoPtr<ov::ITensor> input_ids,
                        "\"NPUW_LLM_MAX_PROMPT_LEN\" or shorten the prompt.");
     }
 
-    process_longrope(m_prefill_request, m_prefill_in_ports, position_ids);
-
     m_llm_profile["1/prefill:1.prepare_for_new_conversation"].record([&]() {
         prepare_for_new_conversation(prompt_length);
     });
@@ -836,6 +834,8 @@ void ov::npuw::LLMInferRequest::infer_prefill(ov::SoPtr<ov::ITensor> input_ids,
         apply_lora();
     });
 
+    process_longrope(m_prefill_request, m_prefill_in_ports, position_ids);
+
     const bool use_chunk_prefill = m_npuw_llm_compiled_model->m_use_chunk_prefill;
     m_llm_profile["1/prefill:3.infer"].record([&]() {
         if (use_chunk_prefill) {