Add DirectLinearSolver support for length-1 vectors

[bmxam]

This PR enables the use of the DirectLinearSolver when the forward function returns vector of length 1. This fixes #199.

Note : I've placed some tests in the examples/4_gallery.jl file. I didn't know exactly where to place them so I thought that maybe, in a futur work, there could be a documentation page where misc. examples are listed, some kind of "gallery". But I can move them elsewhere if you prefer.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/settings.jl	100.00% <ø> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[bmxam]

Oh, I missed the fact that Documenter is trying to build & run all the files in the examples folder. How do you prefer to proceed : should I move 4_gallery.jl elsewhere, or should I tweak make.jl to skip 4_gallery.jl?

[gdalle]

While I like the new example, I wonder whether it brings anything that is not already in the simpler Optim demo, documentation wise? If not, since our main goal is to test the length-1 case, perhaps a simpler test in the test repo would suffice?

[bmxam]

The new example doesn't bring anything new except maybe the computation of a Hessian (only with DirectLinearSolver), but we could also add this Hessian test in the "badsqrt" example.

I've just pushed modifications that remove the "new example" (as well as the 4_gallery.jl file, and add some tests in test/systematic.jl. Tell me if it's ok.

[gdalle]

perfect, thanks! let's leave investigations of second-order stuff for a future PR!