Loads the models declared in application configuration when MLServe starts.

A transient one-shot process rather than logic inside the application callback. Loading models from start/2 would mean a bad model configuration prevents the whole application from booting — including the parts that have nothing to do with inference, and including the health-check endpoint that would tell you why.

In :async mode (the default) it registers each model and returns immediately; models load in the background and report through MLServe.model_status/2. In :sync mode it waits for every model to become ready or fail, which is what you want in tests and short-lived scripts.