Rosenbrock-W method docstrings

[termi-official]

The doc categorization for Rosenbrock-W (after Wolfbrandt) are messed up on the latest release and need to be rolled back before f8f8307 see #2236 (comment) for more discussion and #2236 (comment) for the concrete action to take for the doc strings functions.

[gstein3m]

Maybe Table 1 in https://github.com/hbrs-cse/RosenbrockMethods/blob/main/paper/JuliaPaper.pdf can help in deciding whether it is a W method. Note that the W property usually only refers to ODEs and not to DAEs.
However, the W property does not appear to be utilized in the implementation. The user must then specify an approximated Jacobian matrix themselves.

[ChrisRackauckas]

However, the W property does not appear to be utilized in the implementation.

It's somewhat implicit. If someone uses autodiff=false, the non-W methods can have some deterioration at lower tolerances. Autodiff makes that go away though, which is the most common option.

The best way to use the W property is to come up with some W-reuse scheme between steps, but we haven't really explored that as much as we should have since the best methods have been the Rodas methods, which don't have the property.