fix: increase graph nodes for Megrez-MoE warmup

Megrez-MoE creates many intermediate tensors during MoE FFN construction:
- sigmoid, add, reshape (3x), get_rows, sum_rows, div, view_2d, mul_mat operations
- ggml_top_k internally calls ggml_argsort + ggml_view_4d (2 more tensors per layer)
- Each of 30 MoE layers creates ~35 intermediate tensors during graph construction

During warmup, the graph is built 3 times with different batch sizes, requiring
sufficient memory pool space for all intermediate tensors.

Add 4096 node overhead for LLM_ARCH_MEGREZ_MOE to accommodate these intermediate
tensors (30 layers × 35 tensors/layer ≈ 1050 nodes, doubled for safety margin).

This fixes the 'not enough space in the context's memory pool' error during warmup,
allowing Megrez-MoE to work without the --no-warmup flag.

Tested:
- All 39 tests pass
- Megrez-MoE works with warmup enabled (no crashes)
- Other models (e.g., Gemma-2) are unaffected
- Verified with outputs up to 100 tokens

Author: tamarPal

src/llama-context.cpp

@@ -1382,7 +1382,21 @@ void llama_context::output_reorder() {
 //
 
 uint32_t llama_context::graph_max_nodes() const {
-    return std::max<uint32_t>(1024u, 8u*model.n_tensors());
+    uint32_t base_nodes = std::max<uint32_t>(1024u, 8u*model.n_tensors());
+    
+    // Megrez-MoE creates many intermediate tensors in build_mergez_moe_ffn for each layer:
+    // - sigmoid, add (bias), reshape (3x), get_rows, sum_rows, div, view_2d, mul_mat (per expert)
+    // - ggml_top_k internally calls ggml_argsort + ggml_view_4d (2 more tensors per layer)
+    // Each MoE layer needs ~30-40 intermediate tensors during graph construction
+    // With 30 MoE layers, this adds significant overhead to the graph (30 layers * 35 tensors = ~1050)
+    // During warmup, the graph is built 3 times with different batch sizes
+    if (model.arch == LLM_ARCH_MEGREZ_MOE) {
+        // Add substantial overhead: ~35 intermediate tensors per MoE layer * 30 layers = ~1050 nodes
+        // Double it to 4096 for safety margin during warmup's triple graph construction
+        base_nodes += 4096;
+    }
+    
+    return base_nodes;
 }
 
 llm_graph_result * llama_context::get_gf_res_reserve() const {