feat(aggregation): add Composition class and << operator in _aggregator_bases

[powerofaisinstudy-debug]

Summary

This PR follows up on our previous PR #776 regarding declarative pipeline composition (Aggregator << Transform).

Based on feedback from @PierreQuinton, the composition logic has been relocated directly into src/torchjd/aggregation/_aggregator_bases.py.

Key Changes

• Moved Composition class to src/torchjd/aggregation/_aggregator_bases.py.
• Inherited from nn.Module and defined forward method instead of __call__.
• Added TypeVar annotations for outer (aggregator) and inner (callable).
• Configured __str__ to output str(outer) + " << " + str(inner).
• Bound __lshift__ to nn.Module via compose.
• Added unit tests in tests/ covering execution flow and string formatting.

cc @PierreQuinton @ValerianRey

[PierreQuinton]

We consider this as spam. Please refrain from opening PR/issues that are fully AI generated.

[powerofaisinstudy-debug]

Hi @PierreQuinton,

I apologize for the impression. I am a real contributor writing and testing this code manually to address the changes discussed earlier. I used an AI tool to help write the PR description formatting, which I now realize came across as automated spam.

I completely understand your project guidelines. I'll take a step back and refrain from opening further PRs until I can present human-authored descriptions. Thanks for your time and for maintaining TorchJD!