Switch to in-place `criterion_and_gradient!`

[p-gw]

As noted in #54, a switch to an in-place gradient calculation could speed up performance by avoiding array allocations:

But if in the long run you consider switching to a unified in-place _criterion_andgradient!(grad::Union{AbstractVector, Nothing}, ...), which would skip gradient calculation if grad === nothing, this PR would be irrelevant. Right now the profiling shows that for a 62x24 matrix rotation ~10% is spent in the array allocation (both gradient calculation and projection).

[alyst]

This is a profiling of a varimax rotation of 782x32 matrix:

Looks like in gradient calculation 33% of the time is spent in array allocation, 33% in copying.