Improve criteria performance

[alyst]

Another pass on optimizing criterion performance:

• replace Ones() * M and similar constructs with sum(M, dims=...) -- it should be faster
• replace tr(A' * B) with dot(A, B) where possible -- should be much faster
• more uses of ∇Q (if available) as a temporary matrix for intermediate operations to reduce memory footprint
• simplify and unify the criterion code for CF and Oblimin (now it is more apparent that it's a weighted sum of column and row sums)
• add Equamax and Parsimax on top of the generalized CF/Oblimin criterion code
• TargetRotation: skip weighting if all target matrix elements are specified

[p-gw]

I've added a github action with benchmarks for criterion_and_gradient! in #67. I think they should run if you rebase on main.

[github-actions]

Benchmark Results

	main	00ae4f0...	main/00ae4f044f7dac...
criterion_and_gradient!/FactorRotations.Biquartimin{FactorRotations.Oblique}()	0.572 ± 0.027 μs	0.565 ± 0.3 μs	1.01
criterion_and_gradient!/FactorRotations.CrawfordFerguson{FactorRotations.Oblique, Float64}(0.5)	1.22 ± 0.03 μs	0.396 ± 0.0081 μs	3.09
criterion_and_gradient!/FactorRotations.Geomin{Float64}(0.01)	1.27 ± 4.6 μs	0.932 ± 0.17 μs	1.37
criterion_and_gradient!/FactorRotations.Infomax{FactorRotations.Oblique}()	2.82 ± 2.1 μs	1.39 ± 0.51 μs	2.03
criterion_and_gradient!/FactorRotations.MinimumEntropy()	0.473 ± 0.027 μs	0.457 ± 0.17 μs	1.04
criterion_and_gradient!/FactorRotations.MinimumEntropyRatio()	15.7 μs	16 μs	0.98
criterion_and_gradient!/FactorRotations.Oblimax{FactorRotations.Oblique}()	0.314 ± 0.001 μs	0.315 ± 0.001 μs	0.997
criterion_and_gradient!/FactorRotations.Oblimin{FactorRotations.Oblique, Float64}(0.5)	1.68 ± 0.31 μs	0.389 ± 0.008 μs	4.31
criterion_and_gradient!/FactorRotations.Oblimin{FactorRotations.Orthogonal, Float64}(0.5)	1.66 ± 0.9 μs	0.393 ± 0.0081 μs	4.23
criterion_and_gradient!/FactorRotations.Quartimax()	0.218 ± 0.001 μs	0.217 ± 0.001 μs	1
criterion_and_gradient!/FactorRotations.Simplimax(5)	0.786 ± 0.036 μs	0.595 ± 0.029 μs	1.32
criterion_and_gradient!/FactorRotations.TandemCriterionI()	3.45 ± 1.9 μs	3.14 ± 1.7 μs	1.1
criterion_and_gradient!/FactorRotations.TandemCriterionII()	2.05 ± 0.15 μs	1.58 ± 0.052 μs	1.29
criterion_and_gradient!/FactorRotations.Varimax()	0.156 ± 0.0001 μs	0.155 ± 0.001 μs	1.01
time_to_load	0.54 ± 0.0085 s	0.527 ± 0.0062 s	1.03

[p-gw]

Thanks, looks very good. We can merge this after the minor changes I proposed.

Also, can you add Parsimax and Equamax to the rotation_methods docs for completeness?

closes #27; closes #28

[alyst]

@p-gw Thank you for the review. I've addressed your comments + tweaked the infomax and biquartimin.

[p-gw]

Thanks, very nice PR!