Speed up for classification by refactoring solution interface

[jkosata]

get_single_solution is supposed to be a user-friendly method but is terribly inefficient for internal work, which has been an issue before (#86 , #130 ) . Rather, we want to pass vectors of variables + varied parameters around, in a fixed order.

Here:

• internal constructions of OrderedDict removed (still works as a UI method)
• Jacobians multiple-dispatched to accept both OrderedDict and a vector of variables + varied parameters
• classification refactored to directly call transform_solutions

_classify_default for 1000 solutions of 9 branches, before:

after:

[oameye]

Oeoeo 3x speedup is already very nice! I will review the code today.

[oameye]

i don't see the bug. I think it is better you do it.

[jkosata]

I'm not seeing this, for example with the parametron.jl test, everything runs fine
(there are other issues I'm on though)

[oameye]

I think this is a major improvement! Very nice!

One quick question. What is the idea about batching the threading instead of just using @threads macro of all indices over the solution matrix?

[oameye]

@jkosata I will leave merging up to you. If you want you can tag 0.7.3 in the Project.toml and comment @JuliaRegistrator register in the commit on the master branch (see example).

[oameye]

Before you Tag the master branch to the Julia register, we could also merge the method keyword from branch method-keyword.