Token usage is attributed to one model, not several spellings

Token usage is recorded per model on each conversation. When a model was picked from the dropdown and the provider did not report back which model answered, the usage was filed under the model's node path rather than its identifier — so the same model showed up as two separate rows in the token-usage report, and the row keyed by path could not be priced. Usage is no longer keyed by that path: a model picked from the dropdown now records onto the same row as the model's identifier.