Add Tensor-based implementation of FarthestPointDownSample

[manuelvogel12]

This PR adds a FarthestPointDownSample implementation for tensor-based point clouds. This enables GPU-acceleration for a performance increase of up to 8x. Before, point clouds were copied to the legacy format and then being downsampled on the CPU before being copied back.

• Bug fix (non-breaking change which fixes an issue): Fixes #
• New feature (non-breaking change which adds functionality). Resolves #
• Breaking change (fix or feature that would cause existing functionality to not work as expected) Resolves #

Motivation and Context

FarthestPointDownSample is an important downsampling method, however it's performance render this implementation inapplicable for certain scenarios. This PR is trying to increase the performance

Checklist:

• I have run python util/check_style.py --apply to apply Open3D code style
  to my code.
• This PR changes Open3D behavior or adds new functionality.
  • Both C++ (Doxygen) and Python (Sphinx / Google style) documentation is
    updated accordingly.
  • I have added or updated C++ and / or Python unit tests OR included test
    results (e.g. screenshots or numbers) here.
• I will follow up and update the code if CI fails.
• For fork PRs, I have selected Allow edits from maintainers.

Description

[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.

[manuelvogel12]

I use the legacy method when dealing with CPU-Pointclouds, because my tensor-based implementation is roughly 10x slower on the CPU than the legacy version.

However, with PR #6989, the performance is only 4 times slower (compared to the legacy implementation).

It can be further sped up to 2.5 times slower (compared to the legacy implementation) by replacing

core::Tensor diff = GetPointPositions() - selected;
core::Tensor distances_to_selected = (diff * diff).Sum({1});

with

core::Tensor diff = GetPointPositions() - selected;
diff *= diff;
core::Tensor distances_to_selected = diff.IndexExtract(1,0) + diff.IndexExtract(1,1) + diff.IndexExtract(1,2);

Therefore, I think it makes sense to remove the if (IsCUDA()) condition and also run this version for CPU-Pointclouds, because with the legacy version, properties such as the intensity are lost, and the dtype of other properties might be changed.

[benjaminum]

Thanks @manuelvogel12 for the update!

[benjaminum]

Looks good