Remove normalisation = "all" option from textstat_valence

[kbenoit]

This scores any word not found in the dictionary as 0, in averaging the word values. This doesn't make any sense, because valences may have scales that do not include zero. So while AFINN has 0 in between its -5 to +5 scale, ANEW is scored 1-9. It makes no real sense to normalise over all token counts, since that's the same as assigning non-dictionary matches a zero.

Better to just allow normalisation to be switched off, and users can apply their own normalisations if they prefer. So keep "none" but make the default "dictionary" and 99% of users will want that.