feat: issue-1028, fetch models when user enters api key

[normunds-wipo]

Summary

Solution is intended only for custom configurations. Each time the user enters api key from frontend, it reloads models, including from the endpoint using the new key.

Relevant changes:

• in frontend api layer, update of user key invalidates models, causing to re-load them when next used
• in modelscontroller, get method does not serve from cache anymore, but each time called, retrieves the models
• in loadConfigModels, if api_key is defined as user_provided, check if user has a valid key and use it to fetch models

It might be working also for default endpoints, as now each time key has been entered, the models get refreshed, but this is not tested.

Change Type

• Issue [Enhancement]: if user_provided API KEY value is used, fetch models for user #1028, fetch user models when user provides API_KEY

Testing

Tested just by running code. We are only using custom configuration

Checklist

Please delete any irrelevant options.

• My code adheres to this project's style guidelines
• I have performed a self-review of my own code
• I have commented in any complex areas of my code
• Local unit tests pass with my changes (well tests related to tokenSplit fail with or without my code)

[mkagit]

Is it working without issues @normunds-wipo ?

[normunds-wipo]

Yes. I just merged the main branch and had to adjust for encryption/decryption utility return value change (from string to Promise), but else we are using this code for a couple of months without problems.

[mkagit]

Yes. I just merged the main branch and had to adjust for encryption/decryption utility return value change (from string to Promise), but else we are using this code for a couple of months without problems.

I'll try it. What's your case for the reloading fetch?
I was looking for to, fetch through custom "user_provided" key, models from the LiteLLM virtual keys proxy server.

[normunds-wipo]

The same, query models from LiteLLM. Different users have different keys and potentially different set of models. So we cannot cache all models and need to query models by user. Also once the user enters the key we need to reload models corresponding to the new key. It seems it is doing it correctly.

[dansavu]

This is a useful feature, looking forward to see the PR approved and merged.