Fix appending empty tensor point clouds

[tianrking]

Type

• Bug fix (non-breaking change which fixes an issue): Fixes Merging empty tensor point cloud fails #7091
• 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

A default-constructed tensor PointCloud has no declared attributes. Calling
append() currently reads its missing positions entry before there is a dtype
or shape to validate, so the common incremental-accumulation pattern fails. The
opposite operand order fails as well because the populated cloud expects the
schema-less cloud to provide every attribute.

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

Treat only a point cloud with no declared attributes as the identity for
append() / operator+:

• If exactly one operand has attributes, clone that operand, including its full
  schema, dtype, and device.
• If both operands are schema-less, return a clone that keeps the left operand's
  device.
• Keep the existing validation path for zero-length point clouds that do declare
  a schema.
• Clone the selected operand so the result does not alias its tensors.

The C++ and Python regression coverage exercises both operand orders, schema and
device preservation, the both-empty case, and non-aliasing. The schema-less
operand deliberately uses its default CPU device so accelerator CI also guards
cross-device adoption.

Validation:

• Reproduced the failure with the released open3d==0.19.0: schema-less +
  populated raises an IndexError; populated + schema-less raises the missing
  positions RuntimeError; schema-less + schema-less raises an IndexError.
• cmake --build <build-dir> --target tests -- -j1 (136/136 build steps)
• tests.exe --gtest_filter=PointCloud/PointCloudPermuteDevices.Append/*
  (1 test passed on CPU)
• clang-format --dry-run --Werror on all four changed C++/header files
• YAPF diff check and py_compile for the changed Python test
• pytest --collect-only for python/test/t/geometry/test_pointcloud.py
  (11 tests collected)
• git diff --check

The Python behavior assertions were not executed locally because the Python
binding was not rebuilt; they are included for the project's wheel CI alongside
the directly executed C++ regression.