Add problem_finding pyskill and bare-backtick prompt notice

[ncoop57]

What

Adds a problem_finding pyskill and a prompt-shape hook that lets the user call out process tells with a single character.

• aai_coding/problem_finding.py (new): a skill documenting how to work an open-ended problem so the problem is found before it is solved. Six phases (Understand, Formulate, Question, Alternatives, Implement, Look back), grounded in Getzels & Csikszentmihalyi (1976), Knoblich et al. (1999), Jansson & Smith (1991), Kapur (2020), and Dorst & Cross
  (2001). Includes worked exemplars: the Meno exchange (done well, then done badly with every tell labelled), the matchstick-equation study for Alternatives, and the Lakatos classroom for Look back.
• aai_coding/harness.py: new PF_NOTICE, added to prompt_notices alongside the other prompt-shape notices. A bare backtick prompt asks the AI to name the tells it committed (numbered 1-6) and any user tell it sees (lettered A-F), then rewrite the reply without them.
• pyproject.toml: registers the skill under the pyskills entry points so it appears in list_pyskills().
• tests/test_harness.py: covers the bare backtick and surrounding whitespace firing, and inline backticks not firing.

Why

The tells are easy to state and hard to notice in the moment. The skill lowers the rate; the backtick is the correction when it slips through. It lives in prompt_notices rather than claude_slop because it is about process, not about the writing of the last reply.