FIX: Use raw content when checking for watched words to flag (PR #10104)

This makes flagged watched words match the behavior of censor, requires_approval and blocked watched words.

GitHub

Normally I’d ask for a test but in this case I think it’s probably OK