Implement the core of FW methods as iterators

[matbesancon]

Most methods consist in

1. Setup
2. Iteration until criterion met
3. Cleanup and return

Part 2. could look roughly the same in many algorithms, with the difference being what happens inside each iteration.
For this, an iteration interface could be nice. This also lets the user debug, inspect and so on without us having to anticipate all they might want to inspect at each iteration. The top-level functions can be kept as-is, but users with high-perf will just do the setup and iterations part, without allocating or logging.

Resources:
This discussion mentions the approach in Manopt.jl:
https://discourse.julialang.org/t/ann-optimkit-jl-a-blissfully-ignorant-julia-package-for-gradient-optimization/41063

A blog post describes the iterator approach to solve a linear system:
https://lostella.github.io/2018/07/25/iterative-methods-done-right.html

This is not high-priority, but can make FW more flexible and let users decide what they want to log and how

[matbesancon]

I'll try to find examples in Manopt using it when this becomes relevant