Update dataclass docs

[patrick-kidger]

@jeertmans -- updated the dataclass docs as suggested in #154. In practice I'm recommending equinox.Module rather than chex.dataclass, as it's a bit safer for most purposes + it's part of the same tech stack as jaxtyping / etc.

[jeertmans]

Nice @patrick-kidger! Didn't know the equinox module could help to create JAX-compatible dataclasses. A few options exact (chex, flax, jax-dataclass, and thus equinox), and it is quite hard to decide which one is the "best" (at least the one recommended by JAX authors), but this is not the questions here :-)

Few suggestions:

• as pointed in my comment chore(docs): fix typos in docstrings #154 (comment) in recommend doctests, but you already answered that, so you can ignore ^^';;
• I see you are not using Sphinx but Mkdocs, but a direct to equinox.Module could be nice. But that may also be harder to maintain if the link can be broken (intersphinx mapping helps to prevent that, I don't know about Mkdocs).

Otherwise looks great!

[patrick-kidger]

I'm biased, unsurprisingly, but Equinox is definitely my favourite choice of dataclass. :) It fixes a lot of the edge cases and adds a few features that the others miss: inheritance, bound methods, docstrings, pretty printing, abstract base classes, field converters, ...
Fewer footguns and greater usability are a big part of why I try to mention it where relevant, in place of any of the others.

(You mentioned Chex originally -- between eqx.debug.assert_max_traces and jaxtyping itself, I confess I don't actually Chex any more these days.)

I believe there is a intersphinx-equivalent for the mkdocs stack, but I've not tried digging into it! I think the link should be stable enough that I'm happy to add it directly though; I'll make that tweak now.

Thank you for the feedback!

[jeertmans]

Thanks for the reply! I'll take a look at Equinox then and check if that can replace Chex or not for the use I have.