refactor: add missing features

[sathvikbhagavan]

This PR does:

• Adds full Matrix/Vector{Vector} support (construct, eval, derivative, integral) to Akima, QuadraticSpline, CubicSpline, CubicHermiteSpline, QuinticHermiteSpline, SmoothedConstantInterpolation, BSplineInterpolation/BSplineApprox, and closes remaining integral/derivative gaps in Linear/Quadratic/ConstantInterpolation.
• Adds an analytical integral and Vector{Vector} construction support to SmoothArcLengthInterpolation, and fixes its cache_parameters kwarg, which was silently ignored (always built as if false).
• Fixes a few bugs: broken integral caching for Vector{Vector} data, an integrate_cubic_polynomial array-arithmetic bug, wrong extrapolation indexing in SmoothedConstantInterpolation, a single-point ConstantInterpolation edge case, and a SmoothArcLengthInterpolation aliasing bug where A.(t_vec) (broadcasting) and the documented Vector{Vector}-out in-place form silently returned the same last-evaluated buffer for every element instead of distinct values (root cause: an in_place buffer-reuse kwarg unique to this type, now removed so it behaves like every other interpolation).
• Fixes the Plots.jl recipe, which silently produced garbage (thousands of degenerate one-point series) or errored outright for any Matrix/Vector{Vector}-valued interpolation — plotting effectively never worked for multi-dimensional data before this. Also fixes a related regression risk for BSplineApprox specifically: its "Data points" series must show the true raw data, not the smoothed curve re-evaluated at the knots.
• Fixes show/pretty-printing, which crashed for higher-rank (AbstractArray{T,3}) data and silently mis-rendered Vector{Matrix} data.
• Rewrites SmoothArcLengthInterpolation's _interpolate/_derivative to avoid in-place broadcast assignment, fixing Zygote/Mooncake/ForwardDiff differentiability w.r.t. t (previously failed unconditionally with "Mutating arrays is not supported").
• Adds Zygote/Mooncake AD test coverage for all the new shape-support code paths, plus new t-derivative coverage for SmoothArcLengthInterpolation (its u-gradient through the constructor remains unsupported by both frameworks — documented, not fixed, since it requires rewriting the mutation-heavy geometry-fitting algorithm).
• Documents all of the above in the README (new Feature Support and Automatic Differentiation tables) and docs (SmoothArcLengthInterpolation added to the interpolation listing, tutorial simplified now that Vector{Vector} input no longer needs manual hcat).
• Full test suite passes with zero failures/errors; verified consistent results across shapes, cache_parameters, and all extrapolation modes.

Made with Claude Code

[ChrisRackauckas]

wow, that's a lot 😅 😅 😅 😅 😅 😅 😅 😅 😅 😅 😅 😅 😅

[sathvikbhagavan]

Yeah 😅
I started off with just to add support for higher order arrays for the remaining methods and then went down the rabbit hole. Also, a big chunk of changes are in tests.