Refactor pyccl under an abstract CCLObject.

[nikfilippas]

This is part 2 of the refactoring proposal. It implements a CCLObject which all other base classes inherit from. The aim of this is to be able to centrally control the behavior of every type defined in pyccl, as well as to be able to adopt a uniform framework for object comparison.

The main idea is that the developer only needs to define a __repr__ for the class, and __hash__ and __eq__ will be automatically added. This way, we can check equivalence of any object to any other object, and it is up to the developer to decide the equivalence criteria when writing up the __repr__ method.

Because some objects contain a large amount of data (such as Pk2D - contains huge arrays), hashing is a relatively expensive operation. Therefore, hash and repr for the object are cached the first time they are requested.

But, what if we want to update some parameters? Won't the cached repr be wrong then?

All CCLObjects are immutable by default, unless an update_parameters method is defined. Only through there may the instance mutate, and when it does, the caches are deleted and recomputed on request. Alternatively, there are two ways to mutate an object: either via the UnlockInstance context manager, or via the @unlock_instance decorator.

Docstrings in base.py explain more thoroughly the details of this implementation.

Secondary changes
In an attempt to improve the OOP code structure, I made two more major changes:

1. All template classes are now abstract. So, HaloProfile, Concentration, HaloBias, and MassFunc (and Emulator in PR906) cannot be instantiated because they contain abstract methods.
2. Converted all "special implementation" subclasses to simple functions. These are subclasses that only overload the __init__ method of the parent class, as there is no point that these live as different types. The subclasses converted to functions are (i) CosmologyVanillaLCDM, (ii) all subclasses of the MassDef superclass, (iii) all subclasses of the Tracer superclass. These changes don't affect how users interact with CCL; this is all internal restructure of the codebase.

How to review this
I suggest starting from CCLObject in base.py, then having a look at _repr.py. Everything else relies on these two things.
There are more than 100 changed files, but most of them are in tests/ and benchmarks/, where I've re-written the imports to take advantage of caching and run them faster. This version passes all the tests on my PC and I have been using it for a while, so if you are a developer feel free to install it and play around.

Note: This PR relies on 918, 923, 925, and 933. Hopefully once these are through the review will be much more straightforward as git diff from the default branch will only contain the relevant changes.

---

Branch Dependency Visualization

[damonge]

OK, a few more comments and I'll start with the individual classes and then tests.

[damonge]

Also, have you checked what's happening to the tests?

[nikfilippas]

@damonge tests should be fine now.

[damonge]

Last few comments before I move on to the tests.

[damonge]

Quick comments on tests

[damonge]

Alright, few comments on fancy_repr. Thanks for implementing it.

[damonge]

Alright, final few comments before merging!

[damonge]

OK, can you merge master after the latest GHA branch merge? Then, when all tests pass, I'll approve.

[damonge]

LGTM