Expired deprecations in scipy v1.14 release break river metric

[bdewilde]

Versions

river version: 0.21.2
Python version: 3.11.7
Operating system: macOS 14.4

Describe the bug

When following along with the sentence classification example, the core training loop did not succeed as documented, but instead errored:

File [~/.pyenv/versions/3.11.7/envs/ds/lib/python3.11/site-packages/river/metrics/roc_auc.py:103](http://localhost:8888/lab/tree/~/.pyenv/versions/3.11.7/envs/ds/lib/python3.11/site-packages/river/metrics/roc_auc.py#line=102), in ROCAUC.get(self)
    100     tprs[i] = safe_div(a=tp, b=tp + fn)
    101     fprs[i] = safe_div(a=fp, b=fp + tn)
--> 103 return -integrate.trapz(x=fprs, y=tprs)

AttributeError: module 'scipy.integrate' has no attribute 'trapz'

As far as I can tell, the issue is that the June release of scipy v1.14.0 included a number of "expired deprecations", and integrate.trapz has been removed in favor of integrate.trapezoid,

Steps/code to reproduce

import river.datasets
import river.feature_extraction
import river.metrics
import river.naive_bayes


X_y = river.datasets.SMSSpam()
model = (
    river.feature_extraction.TFIDF(on='body') | 
    river.naive_bayes.BernoulliNB(alpha=0)
)
metric = river.metrics.ROCAUC()
cm = river.metrics.ConfusionMatrix()

for x, y in X_y:
    y_pred = model.predict_one(x)
    if y_pred is not None:
        metric.update(y_pred=y_pred, y_true=y)
        cm.update(y_pred=y_pred, y_true=y)
    model.learn_one(x, y)

[bdewilde]

Oh, nice: Looks like PR #1568 has the fix.