feat: add optional random seed to the Python estimators

[f-dy]

feat: add optional random seed to the Python estimators

What

Adds a seed argument (default -1 = nondeterministic, unchanged behavior) to
the five Python estimators in bindings.cpp:

• findHomography, findFundamentalMatrix, findEssentialMatrix,
  findRigidTransformation, findLine2D

H, mask = pymagsac.findHomography(corrs, w1, h1, w2, h2, seed=42)  # reproducible

Why

There is currently no way to make MAGSAC/MAGSAC++ runs reproducible, which makes
debugging non-deterministic results difficult. A fixed seed gives repeatable
sampling.

How

A small helper sets the seed before the estimator builds its sampler:

if (seed >= 0) {
    gcransac::utils::UniformRandomGenerator<size_t>::setGlobalSeed(seed); // all URG samplers
    std::srand(seed);                                                     // rand()/Eigen::Random
} else
    gcransac::utils::UniformRandomGenerator<size_t>::clearGlobalSeed();

• setGlobalSeed makes all UniformRandomGenerator-based samplers
  (Uniform / PROSAC / Progressive-NAPSAC / Importance) deterministic.
• std::srand(seed) covers the samplers/solvers that use std::rand() /
  Eigen::…::Random() (AdaptiveReorderingSampler, PnP bundle-adjustment,
  DLS/SIFT solvers).
• seed < 0 restores the previous std::random_device behavior.

⚠️ Dependency

This PR depends on danini/graph-cut-ransac#51
("add a process-wide random seed to UniformRandomGenerator"), which introduces
UniformRandomGenerator::setGlobalSeed/clearGlobalSeed.

The submodule bump to the gcransac commit carrying #51 is intentionally not
included in this PR (to keep the diff minimal and avoid pinning a fork commit).
It should be applied once #51 is merged into graph-cut-ransac master. Until
then this branch will not compile against the currently-pinned submodule.

Scope / follow-up

Covers the five estimators present on master. The additional estimators added
in #45 (plane, PnP, radial homography, affine/essential variants) get the same
seed parameter in a follow-up once #45 lands.