metrics(issue-quality): internal surface — distribution, worst-performing issues, drafting feedback

[JSONbored]

Problem

A backfilled score is only useful if it changes how issues get written. That needs a view.

Scope

• Internal (maintainer-only) surface showing: score distribution across issues, trend over
  time, and the worst-performing issues with their component signals so the failure mode is
  legible (ambiguous? under-specified? too large? wrong acceptance criteria?)
• Per-repo and per-drafting-method breakdown, so hand-written and generated issues can be
  compared directly
• Drill-in from a low score to the PRs and decision records that produced it

Acceptance

Someone drafting issues can answer "which of my issues executed badly, and what did they
have in common" without writing a query.

Out of scope

Any public surface, and automated action on the score.

[linear-code]

JSO-7077