Fetcher writes HTML 404 errors instead of skipping broken URLs

[claude-yolo]

Problem

The fetch script is writing HTML 404 error pages to markdown files when platform.claude.com returns 404s.

Evidence

PR #940 contained 250+ broken files (PHP/Terraform API docs) with HTML error content instead of markdown.

Impact

• Pollutes documentation with broken content
• Makes docs repo unusable for those sections
• Search results become meaningless

Solution Needed

Fetcher should:

1. Detect HTML error responses vs markdown content
2. Skip writing files when upstream returns 404/errors
3. Log which URLs failed for debugging
4. Maybe retry mechanism for transient failures

Files Affected

• content/en/api/php/* (125 files)
• content/en/api/terraform/* (125 files)

cc @lroolle - this blocks clean doc updates until fixed