Add AI-generated text detection benchmarks to Evaluation and Benchmarks

[chajahmed]

This matters because of NLP evaluation/classification: AI-vs-human text detection is a binary text-classification task, and detector robustness against adversarial rewrites (paraphrasers, humanizer tools) is an active evaluation problem without a dedicated place on this list yet.

Adds a new "AI-generated text detection:" grouping under Evaluation and Benchmarks with two entries:

• RAID (ACL 2024) - the largest adversarial benchmark for AI-text detectors, already well-established in the literature.
• AI2Human Detection Benchmark - a smaller labeled corpus (raw-AI / human / paraphrase-tool-humanized / LLM-humanizer-humanized) plus a script to score samples against a live detector API. Disclosure: I maintain the AI2Human detector/humanizer product this corpus benchmarks, included alongside RAID as a second reference point rather than the sole entry.

Happy to adjust placement, trim to just RAID, or drop the AI2Human entry entirely if it reads as too self-promotional for the list's bar.