Refactor log-determinant estimation code to work without computation engines

[karlnapf]

We want to delete the files in lib/computation/*

However, we don't want to delete @lambday 's nice code to estimate log-determinants of large sparse matrices.

This task is to re-factor things such that we can delete the computation classes. This is a nice entrance task that does only require to read/refactor existing code and tests. This will give you a nice first exposure to the internals of shogun without being overwhelming. Good non-trivial entrance task therefore.

Here is a rough list of steps
0. Read through the code, coming from the log-determinant examples

1. Compile Shogun withe the log-det code enabled (see guards in code).
   2, Isolate and run all log-det tests
2. Delete all files in shogun/lib/computation. Re-run cmake and make. The build is now broken, i.e. shogun won't compile
3. Fix the build (will require slight API changes)
4. Make sure all tests work
5. Send PR

[shubham808]

@karlnapf i want to take this up.

[shubham808]

After looking at the code I got rid of computation engines and directly computed jobs treating them as independent jobs.
So I deleted engines directory and then debugged to get everything to work.
Is this the right approach ? And should I keep going similarly to eventually get rid of all classes in computation directory ?

[karlnapf]

yes, thats exactly it. make sure to send a pr soon to we can give feedback.
Keep in mind we also want to get rid of independent jobs.
Best would be if you used openmp instead for now

[shubham808]

@karlnapf using omp is a good idea. I got rid of IndependentJob. However, I am stuck at getting rid of JobResult classes also i think keeping them around would simplify things.

[salonirk11]

Also, is it necessary to remove lib/computation/aggregator ? Can we reuse these classes from mathematics/linalg/rattaprox/logdet/computation/aggregator?

[karlnapf]

Everything in the computation folder should go

[karlnapf]

Done via #4103