type stability: use dims=n, not dims=[n]

[stevengj]

Fixes the type-instability (but not the allocations) of #103

[stevengj]

Weird, this fixes the instability when I call mapslices directly, but not via the function:

using Dierckx, Test
import Dierckx: _derivative
spline = ParametricSpline(1:10, rand(2, 10));
x = 5.1; nu = 1; (; t, c, k, bc, wrk) = spline # extract _derivative parameters

# copy of _derivative, with dims=[2] changed to dims=2
my_derivative(t::Vector{Float64}, c::Matrix{Float64}, k::Int, x::Real, nu::Int, bc::Int, wrk::Vector{Float64}) =
    vec(mapslices(v -> _derivative(t, v, k, x, nu, bc, wrk), c, dims=2))

@inferred vec(mapslices(v -> _derivative(t, v, k, x, nu, bc, wrk), c, dims=2)) # succeeds
@inferred my_derivative(spline.t, spline.c, spline.k, x, nu, spline.bc, spline.wrk) # fails

@jakobnissen suggests that @inferred is only checking the vec function in the first case, so mapslices is still unstable.