TimeGEN-1 Azure API key setup and forecast 422 Error

[smdp2000ofc]

@MMenchero
Im part of the Microsoft Finance team and currently trying to integerate the Azure TimeGen API. While integrating with the nixtlar R package, we encountered issues with the validate API and the forecast functionality.

The forecast works as expected when using the Python nixtla package with the same API keys, so that proves we dont have an issue with our deployment.
However, the same functionality fails when using the nixtlar R package, which suggests the issue is specific to the R implementation.

after a deeper dive, seems like the issue issue is around trying to call the api to get model params.

Appreciate this to get resolved at the earliest

[mitokic]

I work with @smdp2000ofc at MSFT. We noticed two issues when using the R package version.

1. When setting the base url with an azure deployment, you need to always add a "/" on the end of the URL. Or else you cannot connect to any of the API endpoints the package calls. This leads to confusion when debugging because copying the URL from Azure ML or Azure AI foundry does not keep the "/" on the end of the base URL. So the user needs to ensure they add it manually each time. That should be handled via a package like fs within nixtlar itself.
2. Another issue with using azure deployed endpoint. There is nothing in the nixtlar docs that states you have to set model = "azureai" within nixtlar::nixtla_client_forecast(). I only found this in the python docs, even then it wasn't specifically spelled out. It was just listed as possible options in a model list.

Please either update your docs to make sure TimeGen-1 azure users don't run into the same issues, or update the code to automatically adjust the base url and model when users want to leverage an azure deployed model. Thanks team!