model_helpers: fail early for empty stimulus set

when given an empty stimulus set, the activations extraction eventually runs in an error where the layer activations are `None` (because there are no stimuli). The code should fail much earlier than this and immediately indicate that it cannot deal with no stimuli.

Locate the model_helpers code responsible for extracting activations and trace how an empty stimulus set reaches the point where layer activations are None. Confirm the earliest appropriate validation point and verify that empty input now produces an immediate, clear failure before activation extraction proceeds.

Written by the indexing model from the issue text.