ENH(sr3): refactor threshold and thresholder property (#548)

As originally written, "threshold" parameter in SR3 classes referred to the l0 norm. 
Appendix A of the SR3 paper included a conversion from l0 threshold to l0 weight,
which was then used in prox, etc.  However, the name threshold stayed around, even
as it was used as a regularizer.  It even applied the l0-specific conversion to l1 and l2
thresholds.

This PR aims to generalize SR3 to other norm by moving threshold-to-weight
calculation to a separate function that the user can call on: 
`calculate_l0_weight()`

examples/16_noise_robustness/utils.py

@@ -1357,10 +1357,10 @@ def hyperparameter_scan_sr3(
     # start with initial guess that all coefs are zero
     optimizer = ps.EnsembleOptimizer(
         opt=ps.SR3(
-            threshold=0,
+            reg_weight_lam=0,
             max_iter=max_iter,
             normalize_columns=normalize_columns,
-            nu=nu,
+            relax_coeff_nu=nu,
         ),
         bagging=True,
         n_models=n_models,
@@ -1418,10 +1418,10 @@ def hyperparameter_scan_sr3(
     for i in range(tol_iter):
         optimizer = ps.EnsembleOptimizer(
             opt=ps.SR3(
-                threshold=tol,
+                reg_weight_lam=tol,
                 max_iter=max_iter,
                 normalize_columns=normalize_columns,
-                nu=nu,
+                relax_coeff_nu=nu,
             ),
             bagging=True,
             n_models=n_models,