[FEAT] Support normalization for static exogenous features

[JuditHalperin]

This PR adds normalization for static exogenous features via local_static_scaler_type in NeuralForecast, extending the existing normalization framework that already supports local_scaler_type for temporal features.

Motivation

Static features can vary widely in magnitude (e.g., a static feature such as building volume, where different buildings might have values of 500, 1,000, and 20,000). Without normalization, these differences can:

• Slow down training convergence
• Cause numerical instability when combined with temporal inputs
• Bias learning toward large-magnitude features

Unlike temporal inputs, statics are constant per entity and do not vary by window, so only local scaling is meaningful.

Implementation

Added new argument:

NeuralForecast(..., local_static_scaler_type="standard")

[CLAassistant]

All committers have signed the CLA.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[marcopeix]

Thanks for the PR, really great work! I have just a small adjustment I would suggest to the docstring. Thanks!

[JuditHalperin]

Hi @marcopeix, all checks passed, but the build-docs check failed due to a permission error. Is there anything I can do?

[marcopeix]

Hi @marcopeix, all checks passed, but the build-docs check failed due to a permission error. Is there anything I can do?

@JuditHalperin , no nothing you can do on your end, this is entirely on us! We will get this sorted out and merged!