[FEATURE] pr-review-toolkit: modernize agents and harden review-pr command

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

The pr-review-toolkit plugin hasn't been updated since initial release. It doesn't use newer agent frontmatter capabilities and the review-pr command lacks the rigor of the code-review plugin.

Proposed Solution

Agent frontmatter:

• Add tools to read-only agents (comment-analyzer, pr-test-analyzer, silent-failure-hunter, type-design-analyzer) -- they shouldn't have Edit/Write
• Add maxTurns to all agents to prevent runaway context burn
• Add effort and initialPrompt to all agents
• Switch read-only agents from model: inherit/model: opus to model: sonnet (quality floor + cost savings)

Portability:

• Remove hardcoded claude-code-internal references from silent-failure-hunter (logForDebugging, logError, errorIds.ts)
• Remove hardcoded TypeScript/React style rules from code-simplifier
• Add "Before You Begin: read CLAUDE.md" sections to comment-analyzer, pr-test-analyzer, and type-design-analyzer (they had no project-standards discovery)

review-pr command:

• Fix allowed-tools: Task (should be Agent), scope Bash to git:* and gh:*
• Add validation pass (each finding re-verified by a separate agent, like code-review does)
• Add false-positive exclusion criteria
• Add --comment flag for posting inline GitHub comments
• Default to parallel agent execution
• Add pre-flight checks (closed PR, draft, duplicate review)

Cleanup:

• Replace "Daisy" references with generic examples
• Replace "Task tool" references with "Agent tool"