GPT integration

[ChristianPavilonis]

What this does
Proxies gpt.js script and additional scripts loaded by gpt.js that do the heavy lifting for ad calls for google.

Change Summary

• Add a new GPT (Google Publisher Tags) integration that proxies the entire GPT script cascade through the publisher's first-party domain, eliminating third-party script loads to securepubads.g.doubleclick.net
• Server-side: Rust module with HTML attribute rewriting, bootstrap script serving (/integrations/gpt/script), and wildcard proxy endpoints (/pagead/, /tag/) that serve Google's scripts verbatim
• Client-side: TypeScript script guard with six interception layers (document.write, src property descriptor, setAttribute, createElement, DOM insertion patches, MutationObserver) to catch and rewrite all dynamically inserted GPT scripts back through the proxy
• Includes a googletag.cmd command queue patch as a hook point for future synthetic ID targeting injection and consent gating
• Adds integration documentation, unit tests (Rust + TypeScript), and default config in trusted-server.toml

Closes #227

Verifying GPT Script Proxying in the Browser
To confirm that GPT scripts are being proxied through the trusted server domain:

1. Open DevTools > Network tab on a page with GPT ads
2. Filter by JS or search for gpt
3. Look for requests to your publisher domain (e.g. publisher.com/integrations/gpt/script and publisher.com/integrations/gpt/pagead/...) instead of securepubads.g.doubleclick.net
4. Proxied responses will include the response headers X-GPT-Proxy: true and X-Script-Source:
5. In the Console tab, look for log messages prefixed with GPT guard: confirming the interception layers are active (e.g. "GPT guard: rewriting document.write script src")

In general, to avoid incomplete hostname regular expressions, any string used to build a regex should be run through a generic “escape for regex literal” function that escapes all regex metacharacters, not only dots. This ensures future additions to GPT_DOMAINS cannot accidentally introduce patterns that match more than the literal hostname.

Concretely, in crates/js/lib/src/integrations/gpt/script_guard.ts, the fallback block in rewriteUrl currently builds the regex with:

new RegExp(`https?://(?:www\\.)?${domain.replace(/\./g, '\\.')}`, 'i')

This manually escapes only dots in domain. Replace this with a helper escapeRegex (defined in this file) that escapes every regex metacharacter: \ ^ $ * + ? . ( ) | { } [ ]. Use that helper both for domain and for any other future regex constructions based on literal strings if needed. The change is localized to this file: add the helper function (near the top, after constants or helpers) and change the new RegExp(...) call to use escapeRegex(domain) instead of domain.replace(/\./g, '\\.'). No behavior changes for current values, but it becomes robust and satisfies the security rule.

[ChristianPavilonis]

We should consider the scope of this integration.

Currently this is doing a lot of rewriting and proxying it's not catching everything but many scripts and 3rd party calls are proxied through the 1st party context.