fix: ask user before proceeding when corpus fits in context window

[sanchaymittal]

Problem

When detect() returns a non-empty warning field (e.g. "Corpus is ~5,005 words - fits in a single context window. You may not need a graph."), the skill's Step 2 action list silently ignored it and fell through to Otherwise: proceed directly — running the full pipeline, dispatching semantic subagents, and burning tokens unnecessarily.

The warning existed in the library output but had no corresponding stop-and-ask logic in the skill.

Fix

Add a check between the large-corpus branch and the Otherwise fallthrough:

- If the `warning` field is non-empty (e.g. corpus fits in a single context window):
  display the warning to the user and ask whether to proceed or answer directly
  without building a graph. Wait for the user's answer before proceeding.

Applied to skill.md and all platform variants (skill-aider.md, skill-claw.md, skill-codex.md, skill-copilot.md, skill-droid.md, skill-kiro.md, skill-opencode.md, skill-trae.md, skill-windows.md).

Test plan

• Run /graphify on a small repo (<10 files, <10k words) — should display the warning and pause for confirmation
• Run /graphify on a large repo (>10k words, no warning) — should proceed as before without prompting
• Confirm all platform variant skill files contain the new check