feat(llama-swap): add ik-llama-cpp package and Qwen3.6-27B MTP config

Add ikawrakow/ik_llama.cpp as a new package with CUDA/Vulkan support,
enabling MTP (Multi-Token Prediction) and IQ4_KS quantization. Wire it
into llama-swap with a new 'ik-qwen3.6-27b-iq4ks-thinking' model config
and 'iq36' alias. Also add a chat template download to the vLLM setup
script and include the binary on lin-va-desktop.

systems/x86_64-linux/lin-va-desktop/default.nix

@@ -9,6 +9,7 @@ let
 
   nvidia-smi = "${config.hardware.nvidia.package.bin}/bin/nvidia-smi";
   llama-cpp = pkgs.reichard.llama-cpp;
+  ik-llama-cpp = pkgs.reichard.ik-llama-cpp;
   stable-diffusion-cpp = pkgs.reichard.stable-diffusion-cpp.override {
     cudaSupport = true;
   };
@@ -129,6 +130,7 @@ in
 
     # Local Packages
     llama-cpp
+    ik-llama-cpp
     stable-diffusion-cpp
   ];
 }