Deterministic registration

[gsakkis]

This PR addresses #1263 for allowing deterministic registration:

• Adds a seed optional parameter to RegistrationRANSACBasedOnCorrespondence and RegistrationRANSACBasedOnFeatureMatching
• Adds a seed member to FastGlobalRegistrationOption
• Replaces the UniformRandInt function with a callable class that accepts an optional seed.

---

This change is 

[update-docs]

Thanks for submitting this pull request! The maintainers of this repository would appreciate if you could update the CHANGELOG.md based on your changes.

[theNded]

Reviewed 8 of 8 files at r1.
Reviewable status: all files reviewed, 2 unresolved discussions (waiting on @gsakkis, @ssheorey, and @yxlao)

---

cpp/open3d/pipelines/registration/FastGlobalRegistration.h, line 73 at r1 (raw file):

                                 double tuple_scale = 0.95,
                                 int maximum_tuple_count = 1000,
                                 unsigned int seed = std::random_device{}())

Also use utility::nullopt here for consistency?

---

cpp/tests/core/Tensor.cpp, line 1702 at r1 (raw file):

    int64_t max_size = *std::max_element(sizes.begin(), sizes.end());
    std::vector<int> vals(max_size);
    utility::UniformRandInt dist_gen(0, 3);

Potentially add a test in registration to verify that the results are deterministic?

[gsakkis]

Addressed all comments, thanks for the review.

[theNded]

Reviewed 1 of 6 files at r2, 6 of 6 files at r3.
Reviewable status: all files reviewed, 6 unresolved discussions (waiting on @gsakkis and @ssheorey)

[dalnoguer]

Hello, it seems that the seed is not exposed in python. Would it be possible to do so?

(Pdb) o3d.pipelines.registration.FastGlobalRegistrationOption(maximum_correspondence_distance=distance_threshold, seed=0)
*** TypeError: __init__(): incompatible constructor arguments. The following argument types are supported:
    1. open3d.cuda.pybind.pipelines.registration.FastGlobalRegistrationOption(arg0: open3d.cuda.pybind.pipelines.registration.FastGlobalRegistrationOption)
    2. open3d.cuda.pybind.pipelines.registration.FastGlobalRegistrationOption(division_factor: float = 1.4, use_absolute_scale: bool = False, decrease_mu: bool = False, maximum_correspondence_distance: float = 0.025, iteration_number: int = 64, tuple_scale: float = 0.95, maximum_tuple_count: int = 1000, tuple_test: bool = True)

Invoked with: kwargs: maximum_correspondence_distance=0.005, seed=0

Thanks!