ci: run the tutorials on PyTorch 2.14

[ducha-aiki]

What does this pull request do?

This switches the tutorials CI to PyTorch 2.14.0 (torchvision 0.29.0), matching kornia's PR CI.

Until now the jobs were labelled torch-2.5.1 but ran something else. kornia/workflows/.github/actions/env@v1.13.0 installed torch==2.5.1+cpu and torchvision==0.20.1+cpu, and then make setup (torch>=2.8.0) replaced both with torch 2.14.1+cu130 / torchvision 0.29.1+cu130 from PyPI. That means a full CUDA download on a CPU runner, and an unpinned torch. From the 2026-10-05 scheduled run:

Pytorch version:  2.5.1+cpu       <- env action
Pytorch version:  2.14.1+cu130    <- make check-deps, after make setup

Changes:

• test job: inlined from kornia/workflows/.github/workflows/tutorials.yml@v1.13.0, with the same steps. The reusable workflow hardcodes extra-deps: torchvision==0.20.1+cpu, which only installs alongside torch 2.5.1, so passing pytorch-version: 2.14.0 to it would hit a resolver conflict. The job now passes pytorch-version and a matching torchvision==0.29.0+cpu to the same env action.
• build job: the same pytorch-version / torchvision pins for the env action.
• requirements.txt: torch>=2.14.0 and torchvision>=0.29.0. With the CPU wheels already installed, make setup keeps them instead of pulling CUDA builds. kornia's own scheduled tutorials job (PIP_INDEX_URL=https://download.pytorch.org/whl/cpu) already resolves to 2.14.1+cpu, so it still satisfies the new floor.

The current daily failure, fit_line.ipynb with negative weights, is fixed separately in #118. This PR's own test run will hit that failure until #118 merges.

What did you check?

uv pip compile requirements.txt requirements-dev.txt + {torch==2.14.0+cpu, torchvision==0.29.0+cpu}
  --python-version 3.12 --python-platform x86_64-manylinux_2_28 (pytorch.org find-links)
  -> resolves: torch 2.14.0+cpu, torchvision 0.29.0+cpu, torchmetrics 1.9.0, pytorch-lightning 2.6.6, no nvidia-* packages
torch-2.14.0+cpu and torchvision-0.29.0+cpu cp312 manylinux_2_28 x86_64 wheels exist on download.pytorch.org
build.yml parses; jobs: test, build, deploy

Not verified: the workflow run itself. That will be this PR's CI.

Posted on behalf of @ducha-aiki by Claude (Opus 5.5).