Add COLNEW algorithm

[ErikQQY]

Add COLNEW method from ODEInterface.jl.

1. The original COLNEW method supports multi-points BVP, its boundary condition must be defined as:
   For example: if in time span $t = [-1, 1]$ we have boundary conditions $u(-1)=1, u'(0)=2.5, u(1)=5$, (here suppose u[1] is $u$, u[2] is $u'$)

zeta = [-1, 0, 1] # points in time span where we want to specify conditions
function bc(i, u, bc)
    if i == 1
        bc[1] = u[1] - 1.0
    elseif i == 2
        bc[1] = u[2] - 2.5
    elseif i == 3
        bc[1] = u[1] - 5.0
    end
end
......
sol, retcode, stats = colnew(tspan, orders, zeta, rhs, Drhs, bc, Dbc, nothing, opt)

the multi-points boundary condition definition in COLNEW makes it hard to convert to the API we are using here, so this PR only implemented the two-point case.

2. The initial guess is set to nothing because we support initial guess as initial guess value(set u0 as Vector) during problem construction in BoundaryValueDiffEq.jl, whereas the initial guess in COLNEW can be either nothing or the returned value from earlier COLNEW call

[github-actions]

Benchmark Results

	master	865d365...	t[master]/t[865d365...]
Simple Pendulum/IIP/MultipleShooting(10, Tsit5; grid_coarsening = false)	1.92 ± 0.66 ms	1.92 ± 0.67 ms	1
Simple Pendulum/IIP/MultipleShooting(10, Tsit5; grid_coarsening = true)	2.64 ± 0.055 ms	2.58 ± 0.061 ms	1.03
Simple Pendulum/IIP/MultipleShooting(100, Tsit5; grid_coarsening = false)	27.4 ± 1.5 ms	27.5 ± 1.4 ms	0.999
Simple Pendulum/IIP/MultipleShooting(100, Tsit5; grid_coarsening = true)	0.898 ± 0.012 s	0.892 ± 0.0065 s	1.01
Simple Pendulum/IIP/Shooting(Tsit5())	0.174 ± 0.0031 ms	0.174 ± 0.0034 ms	1
Simple Pendulum/OOP/MultipleShooting(10, Tsit5; grid_coarsening = false)	3.03 ± 0.11 ms	3 ± 0.11 ms	1.01
Simple Pendulum/OOP/MultipleShooting(10, Tsit5; grid_coarsening = true)	5.56 ± 2.1 ms	5.53 ± 2.1 ms	1.01
Simple Pendulum/OOP/MultipleShooting(100, Tsit5; grid_coarsening = false)	0.0703 ± 0.0015 s	0.0711 ± 0.0016 s	0.989
Simple Pendulum/OOP/MultipleShooting(100, Tsit5; grid_coarsening = true)	2.66 ± 0.046 s	2.68 ± 0.042 s	0.992
Simple Pendulum/OOP/Shooting(Tsit5())	0.685 ± 0.017 ms	0.679 ± 0.017 ms	1.01
time_to_load	4.23 ± 0.026 s	4.23 ± 0.008 s	1

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).

[ChrisRackauckas]

It would be nice to get some kind of side channel for its BC definition just to make benchmarking easier. I assume any translation like that would have to be done manually though, i.e. have inside of the colnew algorithm some bc_func option that overrides the BC with the manual translation

[ErikQQY]

👍Making bc_func inside colnew to override BC with manual translation is a good idea.