[docs] Reference impl of `linear_cross_entropy` probably buggy for multi-dim inputs

### 🐛 Describe the bug

But for BTC input this should probably return a faulty result, as a `transpose(-1, 1)` (or `view(-1, num_classes)` is needed for F.cross_entropy to correctly detect the classes dimension

Also, the docs page even for a single function, as returned by search, includes a redundant `#...` (I would propose to drop these hash components in the url's for single-function html pages): https://docs.pytorch.org/docs/2.14/generated/torch.nn.functional.linear_cross_entropy.html#torch-nn-functional-linear-cross-entropy, cc @svekars @sekyondaMeta @AlannaBurke @ezyang @malfet

### Versions

N/A

Start with the torch.nn.functional.linear_cross_entropy reference implementation and compare its handling of BTC-shaped inputs with F.cross_entropy's expected class dimension. Also inspect the generated single-function documentation URL for the redundant fragment. Done means the reference implementation handles multi-dimensional inputs correctly and the single-function page URL no longer includes the unnecessary hash component.

Written by the indexing model from the issue text.