plain language and report streamline changes for cli-review skill

[protpaw]

Changes

• plain language improvements
• usability improvements (Detailed analysis is a collapsible section)
• added references to skill version, and cli spec (TODO: logic for this is missing)
• streamlined scoring to provide only raw value
• preserved original skill behaviour as much as possible while enabling new template

Validation

• local mock runs (without actually going through Github Actions)
• models used for testing (with VSCode Copilot) : GLM 5.2, Auto, Claude Opus 4.8, Deepseek v4 Pro (2 types: no thinking, medium)
• results scaled predictably with chosen model

TODO

• add logic for new static links
• validate if skill behaviour remained unchanged