BUG: SINDyDerivative set_params() missed sklearn interface

Was resulting in:
```
> self.kwargs
{"kind": "spline", "s": .01, "kwargs":{"kind": finitedifference, "k": 1}}
```

This is because set_params has to mimic the exact functionality of __init__.

See https://scikit-learn.org/stable/developers/develop.html#parameters-and-init

pysindy/differentiation/sindy_derivative.py

@@ -52,7 +52,7 @@ def set_params(self, **params):
             # Simple optimization to gain speed (inspect is slow)
             return self
         else:
-            self.kwargs.update(params)
+            self.kwargs.update(params["kwargs"])
 
         return self
 

test/differentiation/test_differentiation_methods.py

@@ -262,6 +262,13 @@ def test_wrapper_equivalence_with_dxdt(data, derivative_kws):
         )
 
 
+def test_sindy_derivative_kwarg_update():
+    method = SINDyDerivative(kind="spectral", foo=2)
+    method.set_params(kwargs={"kind": "spline", "foo": 1})
+    assert method.kwargs["kind"] == "spline"
+    assert method.kwargs["foo"] == 1
+
+
 @pytest.mark.parametrize(
     "data, derivative_kws",
     [