Add patterns #25-27: persuasive tropes, signposting, fragmented headers

[jacobjmc]

The skill catches the big AI tells but misses a few structural ones I keep
seeing in output. I added three new patterns to cover them.

Pattern #25 is for phrases like "The real question is" and "At its core" that
frame ordinary claims as revelations. The sentence after them almost always
just restates something already said.

Pattern #26 is signposting, things like "Let's dive in" and "Here's what you
need to know." The model announces what it's about to do instead of doing it.

Pattern #27 is subtler. The model puts a short generic sentence right after a
heading ("Speed matters.") before the actual paragraph. It adds nothing the
heading doesn't already say.

I tried to be careful about overlap with existing patterns. Each new one
cross-references the related ones (#25 vs #1 and #22, #26 vs #19 and #22) and
explains the boundary. I also added "Not a problem when" notes to all three
so they don't fire on legitimate uses like op-eds or presentation scripts.

Updated README.md (new table section, count) and WARP.md (count).