ENH: add global meta data

[mj-will]

Based on the discussion in #867, I've made a start on a possible implementation of a global meta data object.

For now, it's just a singleton dictionary that, by default, tracks the rng. Things like the cosmology can then be added when they're set.

Motivation

This change makes it easier to keep track of any global variables and allows us to ensure they're saved in the result file after running a sampler.

Changes (updated 27/01/25)

• Add a new meta_data submodule and GlobalMetaData class
• Add functions for encoding/decoding numpy random generators
• Add test for saving generators
• Extend seed, set_cosmology and _set_default_cosmology to update the global meta data when called.
• Add functions for encoding UserDict and UserList subclasses. These rely on the fact both classes have a data attribute, see the Python docs

Testing

These changes are mostly covered by the existing tests save result files, but I added a test for the encoding/decoding change and for the global meta data class.

[ColmTalbot]

This looks like a good start to me.

[ColmTalbot]

Is using this method instead of just doing global_meta_data[key] = value required, or some convention I haven't come across before?

[mj-will]

I thought we may want to add some form of check/validation when adding values, so this would make it easier in future. Happy to revert to something simpler though

[GregoryAshton]

You can in principle override the get/set item: https://stackoverflow.com/questions/2390827/how-to-properly-subclass-dict-and-override-getitem-setitem

However, my experience when doing this before is that can introduce unexpected behaviour. But, on the other hand it may be a good idea so that any checks are uniformly applied (e.g. if someone utilises the get/set method instead of using add_to_metadata

[mj-will]

That's a good idea @GregoryAshton. I've used UserDict as the base class, since this is less error-prone than inheriting directly from dict. This does mean, however, I've had to add encodings for UserDict (I add UserList as well since they can share the same if statement).

Let me know what you both think

[mj-will]

The tests are going to fail until #867 is merged, since it adds support for properly saving the cosmology objects

[mj-will]

In what seems to be a reoccurring theme, I realised that adding the rng to the result object requires new encoding/decoding functions. For now, I've added them here but I'm happy to move them to another PR.

[ColmTalbot]

This looks good to me now.

[asb5468]

This looks good to me and the tests cover the new functionality suitably.