#3414 - fix _two_points_1d! & better vertices_list of 1D HPolytope

[schillic]

Closes #3414.

Interestingly, Polyhedra's 1D backend is much better. This suggests to use a smarter algorithm for low/high/extrema of HPolytope/HPolyhedron in 1D. I still suggest that we merge this branch and add these algorithms in a separate issue.

julia> P = convert(HPolytope, Interval(1, 2))

julia> @time vertices_list(P)  # new behavior in this branch
  0.000328 seconds (691 allocations: 41.000 KiB)
2-element Vector{Vector{Float64}}:
 [1.0]
 [2.0]

julia> @time vertices_list(P; backend=LazySets.default_polyhedra_backend_1d(Float64))  # old behavior
  0.000092 seconds (51 allocations: 2.094 KiB)
2-element Vector{SVector{1, Float64}}:
 [2.0]
 [1.0]

julia> @time vertices_list(P; backend=LazySets.default_polyhedra_backend_nd(Float64))
glp_simplex: unable to recover undefined or non-optimal solution
  0.001231 seconds (2.28 k allocations: 55.672 KiB)
2-element Vector{Vector{Float64}}:
 [1.0]
 [2.0]

[mforets]

did you compare with @btime? running times can have quite big variance if used only with @time

[schillic]

It was about the allocations mainly. In general this is a non-issue because 1D sets are super fast anyway.

julia> @btime vertices_list(P)
  90.781 μs (691 allocations: 41.00 KiB)

julia> @btime vertices_list(P; backend=LazySets.default_polyhedra_backend_1d(Float64))
  7.331 μs (51 allocations: 2.09 KiB)

julia> @btime vertices_list(P; backend=LazySets.default_polyhedra_backend_nd(Float64))
  354.879 μs (2286 allocations: 55.78 KiB)