DOC: Fix parameter names in private helper docstrings

[karpovantonme]

Ten docstrings name a parameter the function does not take. All in private helpers, so nothing here is user facing, but help() and editor tooltips are what a developer reads when working on this code.

Renames, one to one

Where	Documented	Signature
_lin_field_coeff, _do_lin_field_coeff	rmag, cosmag	rmags, cosmags
write_named_matrix	matkind	mat
_abstract.text2d	x, y	x_window, y_window
_cart_to_sph	cart_pts	cart
_sss_basis	coils	all_coils

The first pair is nice: the line right under it already reads "The sensor assignments for each rmag/cosmag/w", so the singular is the element and the parameter is the plural. write_named_matrix documented matkind, which does not exist, and left the real mat undocumented.

One wrong default

_ajd_pham documents max_iter : int, default 1000, the signature is max_iter=15. That is off by a factor of 66, and it is the kind of number someone reads before deciding whether the algorithm converged. pyRiemann, which the comment right below credits as the source of this implementation, uses n_iter_max=20, so 1000 does not come from there either. I did not touch the code, only the number in the docstring.

Ten more I did not touch

The same sweep flags ten where the parameter was not renamed but removed or split, so the fix is a judgement call rather than a rename:

• _fft_multiply_repeated documents h_fft and n_fft, both now live inside cuda_dict
• _prep_field_computation documents rr and fwd_data, signature is (sensors, bem, n_jobs, verbose)
• _sph_to_cart_partials documents sph_grads, which became five separate arguments
• calculate_lut documents lim_cmap, gone entirely
• plus _fast_sphere_dot_r0 (rr1), _bcd (n_positions), _read_segment (projector), _n_colors (bytes vs bytes_)

Those need someone who knows whether the description should be dropped or rewritten for the new arguments. Point at any of them and I will send a follow-up.

Checks

Docstrings only, no code touched. The sweep reported 19 before and 10 after, and the ten left are exactly the list above.

AI assistance, per the policy: the mismatches were surfaced by a sweep that compares numpydoc Parameters names against the actual signature, run with Claude Opus in Claude Code, which also applied the renames and looked up the pyRiemann default. Each one was checked against the parsed signature before the edit, and I read the diff before submitting. Same setup as #14124 and #14125.

[larsoner]

FYI we have started gradually removing params from many of these docstrings because they tend to get out of date quickly. That said, these seem like pretty straightforward improvements so I'm okay with merging once the changelog issue is addressed

[larsoner]

Accidentally committed?