Replace conda with pyenv to fix incorrect libstdc++ use in jammy CI.

[ssheorey]

Type

• 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

CUDA jammy CI (python tests) fail since pytest causes the conda libstdc++ library to be loaded instead of the system libstdc++ library. This is older and prevents Open3D cuda pybind lib from loading.

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

• Replace python install from conda with pyenv python install. This uses the system libraries.
• Use gcc-12 instead of gcc-11 when building in Ubuntu 22.04. gcc-11 leads to a seg fault in tensormap in Release mode.

[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.

[ssheorey]

Seg fault in python test_tensormap.py here when built with gcc 11 (Ubuntu 20.04 jammy) in release config. No seg fault in debug or relwithdebinfo configs.

...
def test_tensormap(device):
...
    # __delitem__ operator. SEGFAULT
    with pytest.raises(RuntimeError) as excinfo:
        del tm.positions
        assert 'cannot be deleted' in str(excinfo.value)
...
   # Set primary key. SEGFAULT
    with pytest.raises(KeyError) as e:
        tm.primary_key = o3c.Tensor.ones((2, 3), dtype, device)
...
    # Get unknown attributes. SEGFAULT
    with pytest.raises(KeyError) as e:
        normals = tm.normals

Fix: switch to gcc 13.

Update nanoflannimpl to fix this warning as error:

    inlined from ‘open3d::core::nns::impl::{anonymous}::_KnnSearchCPU<float, int, open3d::core::nns::NeighborSearchAllocator<float, int>, 1>(open3d::core::nns::NanoFlannIndexHolderBase*, int64_t*, size_t, const float*, size_t, const float*, size_t, int, bool, bool, open3d::core::nns::NeighborSearchAllocator<float, int>&)::<lambda(const float*, const float*, size_t)>’ at /home/ssheorey/Documents/Open3D/Code/Open3D/cpp/open3d/core/nns/NanoFlannImpl.h:126:5,
    inlined from ‘open3d::core::nns::impl::{anonymous}::_RadiusSearchCPU<float, int, open3d::core::nns::NeighborSearchAllocator<float, int>, 1>(open3d::core::nns::NanoFlannIndexHolderBase*, int64_t*, size_t, const float*, size_t, const float*, size_t, const float*, bool, bool, bool, bool, open3d::core::nns::NeighborSearchAllocator<float, int>&)::<lambda(const tbb::detail::d1::blocked_range<long unsigned int>&)>’ at /home/ssheorey/Documents/Open3D/Code/Open3D/cpp/open3d/core/nns/NanoFlannImpl.h:258:41:
/usr/include/c++/13/bits/new_allocator.h:168:33: error: ‘void operator delete(void*, std::size_t)’ called on pointer ‘<unknown>’ with nonzero offset [4, 9223372036854775804] [-Werror=free-nonheap-object]

[benjaminum]

lgtm

[benjaminum]

Nice catch! This should significantly reduce the number of heap allocations

[benjaminum]

I was working on switching to Miniforge in #6717 😅

[ssheorey]

miniforge likely won't fix this issue, since it's a copy of conda and likely uses it's on libstdc++ as well.

[ssheorey]

Seg fault with Python, CUDA 11.8, 12.1 and Ubuntu 22.04. No issue with C++, the open3d-cpu package, or if CUDA_VISIBLE_DEVICES is empty. No issue on Ubuntu 20.04 or Windows.