Make solver options in LeastCoreValuation discoverable

[janosg]

Current Situation

LeastCoreValuation and its subclasses accept an argument solver_options: dict[str, Any] which contains the name of a cvxpy supported solver and further configuration of the solver. To show which solvers and options are supported, we link to the cvxpy documentation:

• supported solvers
• supported options for each solver

Problems.

• There is no autocomplete support for solvers or solver options
• The name of the same options differs across solvers (e.g. sometimes the maximum number of iteration is called max_iter and sometimes max_iters)

Potential solution 1

We could write a simple abstraction around cvxpy solvers. We can harmonize options for the trivial cases (e.g. max_iter/max_iters) but don't have to go too far for very solver specific options. There are currently 16 relevant solvers in cvxpy.

However, there are drawbacks to this approach:

1. Whenever cvxpy changes, we would have to update the wrappers
2. It seems like such a project should not be part of a data valuation library

Potential solution 2

We could contact the authors of CVXPY and see if they are open to a PR that makes solvers and their supported options more discoverable. While more work initially, the maintenance will become much simpler.

[mdbenito]

I don't think we should wrap cvxpy. At most hint at some good defaults. But I would rank this as low prio