Constrain dev pandas version

[ADBond]

In #2514 I updated the lockfile to get latest package versions, and introduced some minor compatibility fixes. However, something in the newer versions of packages causes problems for pythons >= 3.10. Specifically, poetry's resolver ends up giving us:

• numpy == 2.1.3
• pandas == 2.0.3

Together this leads to ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject.

From what I gather, this is due to a change from numpy == 2.0.0. The first version of pandas compatible with numpys of this major version is pandas 2.2.2. However poetry's resolver is not aware of this fact, as pandas only has lower version constraints (for some versions of pandas). For instance pandas == 2.0.3 does not cap the numpy version, and so as far as the resolver is concerned the above pair of packages are perfectly happy together.

There is no way (afaict) to specify dependent constraints between two packages to restrict this sort of thing. It is also worth noting that with a standard pip install this doesn't seem to be an issue - possibly this is occurring due to the combination of packages we have in dev. At any rate, I am reluctant to include any additional hard constraints on our requirements, as this would be unnecessarily limiting for users, as there are no actual direct breaks for us. Anyone downstream encountering this issue in their own environment can just include appropriate constraints themselves.

Therefore I think the best approach is to introduce a hard constraint in the dev dependencies only. That means for devs + in CI, for pandas we have a minimum version of 2.2.2 whenever python is >=3.10. This means we should once again get a valid set of dependencies, but we don't have to impact users.