chore: bump pi-coding-agent to 0.78.1, reduce llama-swap ctx size, add builder host key

- packages/pi-coding-agent: 0.78.0 → 0.78.1 (new gondolin extension deps)
- llama-swap: reduce qwen3.6-35b-dual macros.ctx 215000 → 131072
- nix: add publicHostKey for remote builder

modules/nixos/nix/default.nix

@@ -38,6 +38,7 @@ in
           sshUser = "evanreichard";
           protocol = "ssh";
           sshKey = config.sops.secrets.builder_ssh_key.path;
+          publicHostKey = "ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIGlpC0rorATKzK8mLMKgCYqM5N2N/GgS2t4M13cwnAOS";
           supportedFeatures = [
             "benchmark"
             "big-parallel"

modules/nixos/services/llama-swap/config.nix

@@ -682,7 +682,10 @@ in
     # https://huggingface.co/unsloth/Qwen3.6-35B-A3B-MTP-GGUF/tree/main
     "qwen3.6-35b-dual" = {
       name = "Qwen3.6 35B (Dual GPU, UD-Q6)";
-      macros.ctx = "215000";
+      # macros.ctx = "215000";
+      # -ctk q8_0 \
+      # -ctv q8_0 \
+      macros.ctx = "131072";
       cmd = ''
         ${llama-cpp}/bin/llama-server \
           --port ''${PORT} \
@@ -694,8 +697,6 @@ in
           --top-k 20 \
           --min-p 0.00 \
           --presence-penalty 0.0 \
-          -ctk q8_0 \
-          -ctv q8_0 \
           --spec-type draft-mtp \
           --spec-draft-n-max 3 \
           -dev CUDA0,CUDA1 \