Remove artificial restriction to refining at least 2 steps.

Refining only one step is now allowed.

src/emd_falsify/path_sampling.py

@@ -1252,7 +1252,7 @@ def generate_path_hierarchical_beta(
        the space between a start and an end point, but it needs the value of _q_
        to be given at those points. These end points are by default Φ=0 and Φ=1,
        which correspond to the bounds of a quantile function.
-    res: Returned paths have length ``2**res``.
+    res: Returned paths have length ``2**res+1``, and therefore ``2**res`` increments.
        Typical values of `res` are 6, 7 and 8, corresponding to paths of length
        64, 128 and 256. Smaller may be useful to accelerate debugging. Larger values
        increase the computation cost with typically negligible improvements in accuracy.
@@ -1276,8 +1276,8 @@ def generate_path_hierarchical_beta(
     #     raise ValueError("`Phi`, `qstar` and `Mvar` must all have "
     #                      "the same shape. Values received have the respective shapes "
     #                      f"{np.shape(Phi)}, {np.shape(qstar)}, {np.shape(Mvar)}")
-    if res <= 1:
-        raise ValueError("`res` must be greater than 1.")
+    if res < 1:
+        raise ValueError("`res` must be greater or equal to 1.")
     rng = np.random.default_rng(rng)  # No-op if `rng` is already a Generator
     # Interpolation
     N  = 2**res + 1