xgboost v2.1.0 is incompatible with pandas<1.2

[winwinashwin]

To Reproduce

import xgboost as xgb
import pandas as pd

X = xgb.DMatrix(pd.DataFrame({"a": [1, 2], "b": [2, 4]}), label=[0, 1])

xgboost==2.1.0, pandas==1.1.5 ---> FAIL
xgboost==2.0.3, pandas==1.1.5 ---> PASS
xgboost==2.1.0, pandas==1.2.0 ---> PASS

---

There is no version constraint specified in pyproject.toml for pandas optional dependency.

The Float32Dtype used here is only available in pandas v1.2+

xgboost/python-package/xgboost/data.py

Line 476 in 213ebf7

from pandas import Float32Dtype, Float64Dtype

[trivialfis]

At the moment, we are not very good at keeping track of these input versions and the policy is always testing with the latest for optional dependencies (pandas is not required to run xgboost). Otherwise, the constraint solver will run into trouble with the amount of input types https://github.com/dmlc/xgboost/blob/master/doc/python/python_intro.rst#support-matrix .

[trivialfis]

Hopefully, #10452 can help in the future.

[hcho3]

For now, let's require Pandas 1.2+. I submitted #10476