msgpack 6.0 changed the cmake config. This PR adds that support.

* infra for tensor index-reduction and interface for voxeldownsample

* basic sum reduction

* temp fix for workload nums

* temp fix with contiguous input

* separate index tensor

* add CPU counterpart

* clean up cpp part

* add unit test for index_add_

* fix point attribute shape

* fix unit tests

* fix doc, also fix several doc issues in t.pointcloud

* fix benchmark

…sl-org#6308)

* SimplifyQuadricDecimation: Check for flipped triangles connected to both vertices of the removed edge

The previous code checks if any triangles connected to the removed vertex are flipped in the process of collapsing the edge. However, since the remaining vertex on the collapsed edge can be moved during the operation, triangles connected to this vertex can flip as well. This commit puts the flip-checking code into a loop so we can check both sets of triangles.

* SimplifyQuadricDecimation: Disallow creating very small triangles

This sets a threshold on how much any triangle can shrink in area during an edge collapse which should avoid creating degenerate triangles. The threshold is somewhat arbitrary, I found 0.001 worked well but it may make sense for this to become a function parameter.

* SimplifyQuadricDecimation: Prevent 'pinching' of triangles

When a vertex is connected to exactly 3 triangles and an edge of one of these triangles not connected to the vertex is collapsed, the remaining 2 triangles end up sharing the same vertices, with opposite normals. This creates a non-manifold location which is nearly always undesirable.
This commit adds a check to see if any of the triangles attached to the collapsing edge share an edge that is not connected to `vidx0` or `vidx1`

* SimplifyQuadricDecimation: Save some memory by clearing some vectors after they are no longer needed

* Update Changelog with SimplifyQuadricDecimation fixes

* SimplifyQuadricDecimation: Reduce memory requirements by removing need for vbars & costs maps

Reduces the memory footprint by 27% in my tests runs ~15% faster as we remove a lot of allocations.

* SimplifyQuadricDecimation: Simplify checking for pinched triangles

Changing from tracking edges (`unordered_set<Vector2i>`) to vertices (`unordered_map<int, int>) gives a good performance increase

* fix 6291

* fix 6236

* address review comments

* Add install with upgrade option when installing latest development package.
* util/install-deps-python.sh was removed in isl-org#2146
* Grammar fix.
* Move open3d_ml and arm to after docker.
* organize TOC - Move C++ and Contribution to end and keep python section together.
* Fix sphinx warnings (Title underline too short).
* Fix extra heading drop when going one level down.
* Fix cross reference links
* Add files/directories to ignore for sphinx, reduces inconsistency warnings.
* Fix `Unknown directive type "code_block"` warning.
* Add uvmaps to official docs, fix - `WARNING: document isn't included in any toctree`.
* Fix yaml lexer name - See https://pygments.org/languages/
* Fix warnings - `Inline substitution_reference start-string without end-string.`
* Only copy open3d_tutorial.py once
* Move list of classes to a global variable
* Use f-strings as requested in PR
* Add python_api_in as ignored directory.
* Fix pybind docs with literal block for tree output
* Fix doxygen docs with code block for tree outpu
* Correct class names in doxygen docs for RedWood dataset

- Update the linux/mac/windows CI to handle python 3.11

…TORCH_OPS

This also require protobuf to be updated to 3.24.0

This is possible because
- tensorflow support is temporarily dropped
- pytorch got updated to 2.0.1

…uild

…sl-org#6320)

…e voxels (isl-org#6325)

Since the triangle can span across multiple voxels, we should not break when we find the first relevant voxel.

Co-authored-by: Sharon Berezalsky <sharon@buildots.com>
Co-authored-by: Rishabh Singh <rishabh.17iitkgp@gmail.com>

…elGrid::RemoveVoxel (isl-org#6023)

Co-authored-by: Rishabh Singh <rishabh.17iitkgp@gmail.com>

…cene (isl-org#6051)

Co-authored-by: Seb Mascha <sebastien@summerrobotics.ai>
Co-authored-by: Benjamin Ummenhofer <benjaminum@gmail.com>
Co-authored-by: Benjamin Ummenhofer <benjamin.ummenhofer@intel.com>

Windows MSVC revert to fmt 6
gcc-12 maybe-uninitialized warning
set VTK compatible version
documentation for draw

Build Open3D viewer Debian package in CI and make available as Github artifact.
Viewer CMake project file from @errissa

Update to C++ 17
TensorFlow 2.13
Tensorboard 2.13
scipy 1.11.2

…n-311

Move PyTorch compile warning to where PyTorch version is available.

Update pip, scipy versions.
Use requirements_test.txt in CI
tensorflow-gpu -> tensorflow for 2.13+
remove CUDA_VERSION setting in ci_utils.sh. THis happens in docker/build_docker.sh now.

Enable users to build wheel with tfops locally with docker container.