fix: add error handling to unsupervised Task.start calls in application.ex

- Grid cache warm-up and ML model load now wrapped in try/rescue
  so silent failures are logged instead of lost

lib/microwaveprop/application.ex

@@ -97,7 +97,15 @@ defmodule Microwaveprop.Application do
     # master: when executed here the loaded rows live on the app master's
     # heap and never GC (it's idle after boot), pinning ~800 MiB for the
     # lifetime of the pod.
-    {:ok, _pid} = Task.start(fn -> Microwaveprop.Weather.warm_grid_cache_from_latest_profile() end)
+    {:ok, _pid} =
+      Task.start(fn ->
+        try do
+          Microwaveprop.Weather.warm_grid_cache_from_latest_profile()
+        rescue
+          e ->
+            Logger.error("Application: grid cache warm failed: #{Exception.format(:error, e, __STACKTRACE__)}")
+        end
+      end)
 
     # One-shot backfill: enqueue a QRZ-lookup job for every user with a
     # missing home QTH. Idempotent — the worker no-ops if home_grid is
@@ -118,7 +126,12 @@ defmodule Microwaveprop.Application do
     if Application.get_env(:microwaveprop, :load_ml_model, false) do
       {:ok, _pid} =
         Task.start(fn ->
+          try do
             Microwaveprop.Propagation.load_ml_model()
+          rescue
+            e ->
+              Logger.error("Application: ML model load failed: #{Exception.format(:error, e, __STACKTRACE__)}")
+          end
         end)
     end