docs: support source-repository skill audits

[Sun-sunshine06]

Source-repository skill audits already show up in real use: docs-first skill repositories often have strong validators, CI, example prompts, and validation logs, but still lack clean installed-skill routing transcripts. The previous wording made that gap read too much like an automatic P0 and also assumed a narrower shell/platform story than these repositories actually have.

This PR updates skill-optimizer to handle that case more honestly and more usefully.

What changed:

• add source-repository audit guidance to the README (EN/ZH)
• add severity calibration so missing routing transcripts become P0 only when they contradict the repo's public claims or release gate
• clarify Codex invocation evidence vs skill loading
• document Windows/PowerShell-friendly audit expectations
• split the skill body into a leaner SKILL.md plus references for session analysis, audit framework, report template, and a repo-mode example
• add a fuller README example output that shows the intended report shape in source-repository mode

Why:

• docs-first and beta skill repositories need audits that can say "strong static quality, weak routing proof" without overstating the severity
• short action anchors in descriptions are sometimes boundary markers, not harmful workflow leakage
• the optimizer should model progressive disclosure itself if it recommends that pattern to others

Validation:

• git diff --check
• reviewed the new README example output and reference files for consistency with the updated SKILL.md