AiProviderClientBase::getApiDefinition performance concerns

This needs to be fixed as discussed on Slack - I have added this in the reflection method we have of the providers.

Its currently just being used by getAvailableConfiguration method and could we cached there, but there is no guarantee that it stays like this, thus the reflection method.

Another option is to change this in each provider integration and let them take decisions if it should be cached, but since its the default provider layers configuration it should not be dynamic anyway. The models specific settings might be dynamic, but this can still be 100% controlled by the provider.