Fix/random state, solving #29

[gAldeia]

This pull request improves the determinism and reproducibility of the Brush library by ensuring that a fixed random seed produces identical results across runs, regardless of thread scheduling. The changes affect both the C++ and Python code, update the handling of random number generation, and add comprehensive tests to verify reproducibility.
Deterministic Randomness and Reproducibility

• The C++ engine now executes fitness evaluation and selection for each island in a deterministic order, instead of relying on thread scheduling, ensuring reproducible results when a fixed random seed is set.
• The random number generator (Rnd) now resets its internal Box-Muller cache state (has_spare_normal, spare_normal) when seeded, and its Gaussian sampling (gasdev) implementation is made thread-safe and deterministic.
• The Python estimator seeds the native (C++) random generator at the start of fitting if a fixed random_state is provided, making Python-side runs reproducible as well.
• The documentation for random_state is updated to clarify that a fixed integer ensures full reproducibility, including for multi-island runs.

Testing and Validation

• Extensive tests are added or updated in the C++ test suite to set a fixed random seed (params.set_random_state(42)) in all relevant places, and a new test (FixedSeedProducesIdenticalRuns) asserts that repeated runs with the same seed produce identical models, weights, and populations.

[gAldeia]

Looks good