Improve error handling of user input and API responses

[magicznyleszek]

Description

Ideally this should be done on feature/nlp branch. We should probably introduce some onBackendError (handleBackendError?) utility method that should be imported and used everywhere we want to observe API call failures. It could be passed directly or if we require some more handling in some cases, (response) => {onBackendError(response); <more code>}.

A card from NLP QA board says:

single processing view shows {'detail': 'whatever'} in errors instead of unwrapping the detail

to reproduce: log out in another tab and click "begin" on the transcript screen.

additional: if you stop your kpi (python) server and click "begin", you'll see a toast notification with the red X indicator but no error text.

Additional details

Zulip topic

[jnm]

I added a quick workaround (@dorey's) on the 3939-knowncols+qa-fixes branch just to prepare for the imminent NLP deployment:
5b3bb0c

There's going to be a lot more NLP QA work after the initial, invitation-only release, so this can continue on top of feature/nlp (or just on top of beta if that's easier). Thanks 👍

[magicznyleszek]

feature/regional_projects introduces a function named handleApiFail https://github.com/kobotoolbox/kpi/blame/86e8e14f5a8591fff5ca84b67211eb5a5eaae2a8/jsapp/js/utils.ts#L75