Paper/README updates to match code.

[sidxz]

I am consolidating text-level things I encountered while checking the paper's claims against the code.
None of them are big, just needs updating.

1. Table 1 and the sentence "All handle multidimensional data via axis" (and the README's "All methods can accept blocks of multidimensional data") include lineardiff, but it doesn't have an axis argument:

   lineardiff(np.stack([x, x]), dt, order=3, gamma=0.01, window_size=11, solver='CLARABEL', axis=1)
   # TypeError: lineardiff() got an unexpected keyword argument 'axis'

   Your savgoldiff uses axis 0 length when differentiating another axis #211 commit message already says "the other eleven axis-aware methods", so I guess the text just didn't get updated.

2. State of the Field says the derivative package "lacks multidimensional support", but derivative.dxdt(x, t, kind, axis=0) does have an axis argument (and takes a t array). The NaN-handling and optimizer parts of that sentence look right to me.

3. Affiliations: Please cross check that the authors match their respective affiliations, as mentioned in their ORCID.

Review reference: openjournals/joss-reviews#11172

[pavelkomarov]

1. I actually just forgot to generalize lineardiff, because it's been like an unwanted stepchild the whole time, prone to cause convex solver troubles, yielding poor results, and complicated to look at in code. It's the one method I didn't get knee deep in and entirely rewrite. It's excluded from the analysis in Notebook 4 for these reasons. I'll generalize it and try to clean it up if I can, maybe try to lengthen some windows or something and see whether I can get it to behave more stably.
2. That's what I get for using Claude. I know the derivative package does multidimensional data, because this is critical for its use in the pysindy package (although derivative doesn't do everything pysindy wants, so pysindy also has a local differentiation module for certain things). Another collaborator (@Jacob-Stevens-Haas) opened very old issues on derivative and pynumdiff about possible duplication between the packages. Even before all the updates he thought pynumdiff was the more method-complete suite, and we've talked about making pysindy dependent on pynumdiff now that it's more feature-complete and has really pulled ahead. Multidimensional data handling was always meant to support that effort: Depending on PyNumDiff for differentiation: reduce maintenance burden, gain richer methods dynamicslab/pysindy#688
3. The affiliations are correct. Kutz took an industry job and moved to London, where I presently am as well for a year, and Maria was an undergrad in Kutz's primary department, so AI keeps suggesting I must have flipped them, but I didn't. It is odd that they're numbered out of order, though. I'll fix that by hand shortly.