Improve rosomaxa algorithm

[reinterpretcat]

The general goal of this ticket is to have more stable search result, possible improvement areas:

• GSOM implementation: experiment with smoothing phase, improve growing phase
• tweak rosomaxa intermal logic (TBD)

Basic idea is to extract rosomaxa as separate crate and make it work on non VRP related data. Then it should be possible to use various metaheuristic benchmarks (e.g. Rosenbrock function) to check its quality. Requires some work and research on how to do such benchmarking. Additionally, it would be nice to have some visualizations (?).

[reinterpretcat]

that was mostly done in multiple iterations