Fixes #26808

Speed up confusion_matrix calculation.

This achieves the speedup by calculating the unique values on y_true and y_pred only once, and adds the required args along the way to the inner methods.

Not sure about the names of the added args on public functions, and might need to refactor to hide them if necessary.

Also this needs to add the same for some other functions to make the tests pass, but for now this works, and brings the time from about 15s to 5s on my machine.

import numpy as np

from sklearn.metrics import classification_report

y_true = np.random.randint(0, 2, size=2**23)
y_pred = y_true.copy()
np.random.shuffle(y_pred[2**20 : 2**21])

print(
    classification_report(
        y_true=y_true,
        y_pred=y_pred,
        digits=10,
        output_dict=False,
        zero_division=0,
    )
)