Update paper: MNE wrapper, native engine, Newton note

[neuromechanist]

Updates paper.md (JOSS) for the recent additions, kept concise (body 1748 words, under the 1750 target), with detail pushed to the docs.

paper.md

• MNE compatibility: corrected the now-false "does not yet ship an MNE-Python wrapper" (0.3.0 added AMICAICA); it now notes the wrapper ships as an optional extra.
• Native Fortran engine: added that pamica can run the reference Fortran itself from Python via a bundled dependency-free native build (no MKL or MPI runtime).
• Newton note: added a concise, honest note on why the parity comparison uses do_newton=0 — with Newton on, the backends' distinct RNGs mean a fixed seed is not a matched start, so long runs can settle a few under-determined components in different equally-likely optima, while a matched initialization agrees.

Offsetting trims to verbose passages (permutation-test wording, native-Fortran caveat, threshold sentence, table caption, redundant "documentation" pointers) keep all facts and stay under budget.

docs/guides/validation.md

• Added a "Newton-enabled runs and the initialization basin" subsection (the paper's "(documentation)" complement): the same-init agreement (~0.997, no collapse), the RNG/seed distinction, and the init-basin framing (issue AMICATorchNG diverges from Fortran on weak components at 2000 iters, despite Newton making the optimum sharply unique #145, follow-up Improve NG init-robustness on under-determined components (best-of-N restarts) #198).