Add AgentPipe (cross-vendor adversarial review pipeline)

[Xiaofuziod]

Adds AgentPipe to Autonomous Loop Runners.

AgentPipe implements the keep-running-until-done pattern with a twist: the writer and the reviewer are different vendors on purpose — Claude writes, Codex reviews — and the loop repeats until Codex reports clean. Verify gates can be by: command (your own test/build/lint; exit code 0 = goal met), and convergence is fail-closed (a missing verdict or hitting the loop max stops for a human, never passes silently). Rust engine + Tauri desktop GUI; macOS-first; MIT.

Entry follows the section's [name](link) - description format. Thanks for maintaining the list!

[Keesan12]

This PR highlights a category boundary the list will probably need more often soon.

AgentPipe is not just "another thing that loops" - the interesting part is the adversarial review boundary between writer and reviewer. That's a different design choice than a plain keep-running-until-done runner.

A list structure that separates:

• execution loops
• review / challenge pipelines
• trust / governance / evidence layers
  would make discovery much sharper for builders trying to solve a specific failure mode.

As more projects show up, the taxonomy itself becomes part of the value of the list.