fit should recreate the solver for new data

[kjgm]

Currently, fit reuses the persistent solver when refitting. This is okay when fitting with the same data, but with a higher node budget, since the cache can be reused.
But if the data is different, then the cache is no longer valid, hence the solver should be re-initialized.

Example to reproduce the error:

import pymurtree
import numpy
from sklearn.tree import DecisionTreeClassifier

# Create training data
x1 = numpy.array([[0, 1, 0, 1], [1, 0, 0, 1], [1, 1, 0, 0]]) 
x2 = numpy.array([[1, 0, 1, 0], [0, 1, 1, 0], [0, 0, 1, 1]]) 
y = numpy.array([5, 5, 4]) # labels

model = pymurtree.OptimalDecisionTreeClassifier(max_depth=4, verbose=False)
#model = DecisionTreeClassifier(max_depth=4) # As a reference, when using the sklearn DT classifier, this does work

# fit on first data set
model.fit(x1, y)
labels = model.predict(x1)
assert(all(labels == y))
print("First tree computed perfectly")

# fit on second data set
model.fit(x2, y)

# Predict labels for a new set of features
labels = model.predict(x2)
assert(all(labels == y))
print("Second tree computed perfectly")