Fix weighted_average reduction for dim=0

[Robertboy18]

Issue #, if available:

None.

Description of changes:

Hi! I was playing around with formalizing the behavior of weighted_average in Lean with TorchLean, and noticed that dim=0 takes the all-dimensions branch because zero is falsey.

When weights are provided, this collapses the whole tensor to a scalar instead of reducing along the first axis. This changes the condition to distinguish dim=0 from dim=None and adds a regression test with nonuniform weights. The existing all-dimensions behavior for dim=None stays the same.

Tests:

• PYTHONPATH=src python -m pytest test/torch/test_torch_util.py -q (8 passed)
• ruff format --check src/gluonts/torch/util.py test/torch/test_torch_util.py
• ruff check src/gluonts/torch/util.py test/torch/test_torch_util.py

I also have a corresponding TorchLean tensor formalization of the axis-zero
shape and value if that would be useful :)

By submitting this pull request, I confirm that you can use, modify, copy, and
redistribute this contribution, under the terms of your choice.

Suggested label: bug fix

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.