Fix RT-DETR indexing error when num_feature_levels exceeds backbone o…

[c1prk]

What does this PR do?

Fixes a TypeError in RTDetrModel.forward that occurs when num_feature_levels is set higher than the number of backbone output levels (default is 3).

When extra decoder feature levels need to be synthesized, the first extra level passed the entire list of encoder feature maps to a Conv2d layer and indexed [-1] on the result, instead of indexing [-1] on the input list first to select a single tensor. The corrected line now matches the pattern already used correctly on the very next line of the same block.

# Before (buggy): indexes the conv2d output, passing a list into conv2d
sources.append(self.decoder_input_proj[_len_sources](encoder_outputs.last_hidden_state)[-1])

# After (correct): indexes the input list first, passing a single tensor
sources.append(self.decoder_input_proj[_len_sources](encoder_outputs.last_hidden_state[-1]))

The change is made in modular_rt_detr.py (the source of truth) and modeling_rt_detr.py is regenerated from it. The default config (num_feature_levels=3) is unaffected, so no existing behavior changes. A regression test was added that builds the model with num_feature_levels=4 and confirms the forward pass returns logits of the expected shape.

AI assistance was used to help diagnose and draft this fix; I reviewed and verified every changed line and reproduced both the bug and the fix myself.

Fixes #46832

Code Agent Policy

• I confirm that this is not a pure code agent PR.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline and the Pull Request checks?
• Was this discussed/approved via a Github issue or the forum? [RT-DETR] TypeError in RTDetrModel.forward when num_feature_levels > len(backbone outputs) #46832
• Did you make sure to update the documentation with your changes according to the guidelines?
• Did you write any new necessary tests?

Test results

python -m pytest tests/models/rt_detr/test_modeling_rt_detr.py -k num_feature_levels -v

(passes locally)

Who can review?

@yonigozlan @molbap

[github-actions]

CI Dashboard: View test results in Grafana

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: d_fine, pp_doclayout_v2, rt_detr, rt_detr_v2