[Widget] ConversationalWidget: Better UX / error handling when inference API throws a 503

[SBrandeis]

Right now if the underlying model is re-deploying, the widget just looks broken / unresponsive

The network call does end up in a 503 though - we should display an error message

[pcuenca]

An error does come up after it timeouts. Maybe a progress indicator of some sort would be enough?

[omjeem]

We can utilize the Toastify library to enhance the current progress with customizable and visually appealing toast notifications.

[coyotte508]

We can utilize the Toastify library to enhance the current progress with customizable and visually appealing toast notifications.

We try to avoid adding external libraries in the non-dev dependencies, but we could take inspiration from it