fix: kNN zero-variance dims get inf std; hot-start SAC from saved model

- nucon/model.py: constant input dimensions (zero variance in training
  data) now get std=inf so they contribute 0 to normalised kNN distance
  instead of causing catastrophic OOD from tiny float epsilon
- scripts/train_sac.py: add --load, --steps, --out CLI args; --load
  hot-starts actor/critic weights from a previous run (learning_starts=0)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

nucon/model.py

@@ -95,7 +95,10 @@ class ReactorKNNModel:
         self._raw_states = np.array(raw)
         self._rates      = np.array(rates)
         self._mean = self._raw_states.mean(axis=0)
-        self._std  = self._raw_states.std(axis=0) + 1e-8
+        raw_std = self._raw_states.std(axis=0)
+        # Dimensions with zero variance in the training data carry no distance information.
+        # Use inf so they contribute 0 to normalised L2 (i.e., are ignored in kNN lookup).
+        self._std  = np.where(raw_std < 1e-6, np.inf, raw_std)
         self._states = (self._raw_states - self._mean) / self._std
 
     def _lookup(self, s: np.ndarray):