docs(skills): improve glab threaded-reply guidance

What does this MR do and why?

Improves the bundled glab skill so coding agents post threaded replies (and other rich comment bodies) correctly on the first try.

What we saw

Debugging Duo Developer, we kept hitting a case where it left throwaway "test" comments when asked to reply in a thread with a markdown-heavy body — one that starts with @ and has backticks, $, and emoji. The probe is a symptom: the agent inlines the body into glab api … -f body="…" / -F body=@…, hits shell-quoting errors (and -F reads a leading @ as a filename), falls into trial-and-error, and posts a "test" comment to check the call — which its token can't delete, so it leaks into the thread.

We also noticed that when the agent actually reads this skill it posts cleanly; when it doesn't, it flails. So the goal is to make the skill both easier to reach for and safer to follow.

Full analysis and traces: ai-assist#2386 (findings here).

Changes

• Description — reworded so the skill reads as the reference for any GitLab operation (read or write), not just a subset, so agents consult it before acting. The resource list is explicitly illustrative now.
• Issue / work-item threaded replies — lead with the write-the-body-to-a-file then -F body=@file pattern instead of inlining rich markdown (which breaks on backticks, $, newlines, and a leading @). Short inline form is still noted for trivial bodies.
• Content-type guidance — explains why a body starting with @ must not go through -F (it is read as a filename).