Docs CI is broken

[scotts]

🐛 Describe the bug

Our CI for docs is broken. See example: https://github.com/pytorch/torchcodec/actions/runs/10744233389/job/29800698755?pr=202. The failure is:

generating gallery for generated_examples... [100%] basic_example.py
Warning, treated as error:

../../examples/basic_example.py unexpectedly failed to execute correctly:

    Traceback (most recent call last):
      File "/home/runner/work/torchcodec/torchcodec/examples/basic_example.py", line 66, in <module>
        decoder = SimpleVideoDecoder(raw_video_bytes)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
      File "/home/runner/work/torchcodec/torchcodec/src/torchcodec/decoders/_simple_video_decoder.py", line 1[31](https://github.com/pytorch/torchcodec/actions/runs/10743607733/job/29798699894?pr=201#step:8:32), in __init__
        core.scan_all_streams_to_update_metadata(self._decoder)
      File "/usr/share/miniconda3/envs/test/lib/python3.12/site-packages/torch/_ops.py", line 667, in __call__
        return self_._op(*args, **kwargs)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^
    NotImplementedError: Could not run 'torchcodec_ns::scan_all_streams_to_update_metadata' with arguments from the 'CPU' backend. This could be because the operator doesn't exist for this backend, or was omitted during the selective/custom build process (if using custom build). If you are a Facebook employee using PyTorch on mobile, please visit https://fburl.com/ptmfixes for possible resolutions. 'torchcodec_ns::scan_all_streams_to_update_metadata' is only available for these backends: [CUDA, Meta, BackendSelect, Python, FuncTorchDynamicLayerBackMode, Functionalize, Named, Conjugate, Negative, ZeroTensor, ADInplaceOrView, AutogradOther, AutogradCPU, AutogradCUDA, AutogradXLA, AutogradMPS, AutogradXPU, AutogradHPU, AutogradLazy, AutogradMeta, Tracer, AutocastCPU, AutocastXPU, AutocastCUDA, FuncTorchBatched, BatchedNestedTensor, FuncTorchVmapMode, Batched, VmapMode, FuncTorchGradWrapper, PythonTLSSnapshot, FuncTorchDynamicLayerFrontMode, PreDispatch, PythonDispatcher].

The PR that kicked off this job only added a single space to the README: #202. This code being reference has not changed recently.

The only idea I have is that we're pulling PyTorch nightly: https://github.com/pytorch/torchcodec/blob/2b4f8094e1bba2ff30225c70209b25d60fc2bca8/.github/workflows/docs.yaml#L31
Could something have changed upstream with how these operators get registered? However, the same is true for other tests that succeed. For example: https://github.com/pytorch/torchcodec/blob/2b4f8094e1bba2ff30225c70209b25d60fc2bca8/.github/workflows/wheel.yaml#L81

Versions

Internal CI.

[NicolasHug]

The doc job run on 3.12 while our test run on 3.9 so I suspected the python version might be a factor, but it turns out not to be the case. In #204 it's clear that the doc job is failing on all Python versions and our unittests are passing for all Python versions as well.
The torch version these jobs are using is also the same.

The only difference I see is that the doc job builds from source while the unittests jobs install torchcodec from a wheel. Other than that, I can't think of anything.

Sometimes, these things just... resolve on their own :/

EDIT: in any case, it is very likely to be an upstream issue.

[scotts]

Should be fixed by #218. As mentioned in the comment added in that PR, we don't know the root cause, but we want to fix the doc build now.