Add LLM Benchmark Costco for benchmark discovery and protocol inspection

[joe1chief]

Adds LLM Benchmark Costco to §6 as a resource for inspecting task-construction and grading protocols before choosing a benchmark to reproduce.

I maintain the project. The browser brings together benchmark metadata, English/Chinese summaries, original-source links, and separate construction/evaluation flowcharts. The entry identifies it as a benchmark browser and includes a source-verification caveat.

Try the AutomationBench example, then open Build Process. It separates task construction from evaluation and distinguishes public-task local results from the official private leaderboard. The browser helps readers locate protocol details; it does not run evaluations or produce model rankings.

For the inclusion-evidence requirement: as of September 7, 2026, the repository has 6 stars, from five accounts other than mine plus my own. The repository was created on March 2, 2026. This is an early external-interest signal under the contribution guide's star criterion, not a claim of verified production adoption. I do not yet have an independently documented deployment or user case study to cite.

Validation: this changes only one README entry; checked the project, live browser, and example links, and ran git diff --check.