[ENH] Add DL anomaly detection algorithms: USAD

[SebastianSchmidl]

Describe the feature or idea you want to propose

Audibert et al. proposed a DL-based anomaly detection algorithm for multivariate time series in https://doi.org/10.1145/3394486.3403392. We should include this algorithm in aeon.

There is currently no corresponding implementation in TimeEval. A community implementation can be found here: https://github.com/manigalati/usad

Describe your proposed solution

1. Come up with an API design for deep-learning based anomaly detection algorithms
2. Implement USAD algorithm in aeon

Describe alternatives you've considered, if relevant

No response

Additional context

No response

[aryanpola]

@aeon-actions-bot assign @aryanpola

[Sharathchenna]

@aeon-actions-bot assign @Sharathchenna

[SebastianSchmidl]

This issue depends on the design of the AD DL network architecture, which we have not finalized yet. If you have suggestions/requirements for this design, please participate in the discussion in #2125.