checkpoints.restore_checkpoint` should raise an error for explicit paths that don't exist

^{Originally posted by **PgLoLo** October 9, 2021}
From documentation of `flax.training.checkpoints.restore_checkpoint`:
>Returns:
> Restored `target` updated from checkpoint file, or if no step specified and
> no checkpoint files present, returns the passed-in `target` unchanged.
> If a file path is specified and is not found, the passed-in `target` will be
> returned. This is to match the behavior of the case where a directory path
> is specified but the directory has not yet been created.

Why silently hiding the checkpoint absence (which is a great potential for a bug: by path misspecification, step misspecification, etc.) is a good idea? Makes no sense to me, is there any logic behind this decision?