Fix non-f-string ValueError in t5gemma _normalize_token

[PratikDhanave]

In gemma/research/t5gemma/sampling.py, _normalize_token raises:

raise ValueError(
    'Invalid forbidden token: {token!r}. Forbidden tokens must map to'
    ' single token ids in the vocab.'
)

The first line uses {token!r} but is not an f-string, so the placeholder is emitted literally — the user sees Invalid forbidden token: {token!r} instead of the actual offending token. Add the f prefix so the token is interpolated. (The second concatenated line has no placeholders and correctly stays a plain string.)