feat: add NOAA and USGS environmental observation pilots

[yangziao56]

Summary

Adds an isolated test-cases/environmental-observation corpus with two pilots: NOAA Boston observed water levels and USGS Charles River daily mean discharge. Related to #375; submitted following the maintainer's invitation. Released V1/V2 are unchanged.

The evaluated endpoint is a correctly configured data request, not a saved CSV/ZIP. The matchers distinguish observations from predictions, daily mean from continuous/field data, exact dates and units, and reject duplicate parameters and extra result filters. No site API-key value is included.

Validation

• Both task files pass the current repository task schema; diff whitespace check passes.
• Previously validated against official runtime commit 9dd9d442c8ec720ba4d6d4c95f2bbc0ff2a2d0df: 87 offline request checks, eight real Chromium/CDP request cases, and each complete website UI path through request initiation in a clean browser container.
• NOAA's 480 observed values/timestamps matched a native-browser export; USGS returned all 365 distinct dates with the correct station/statistic/unit metadata.
• These are Codex-driven checks, not human timing studies. Full runner CLI and Stage-2 LLM judge were not run; no paid model run is claimed. The README states these limits and review steps.

Review scope

Please confirm this request-initiation outcome and corpus layout are appropriate. I am submitting the two working pilots for review before the proposed 12 distinct workflows. The category recognition and individual co-authorship route for a particular follow-up paper remain unconfirmed; this pilot does not claim that threshold is met. If the desired outcome is instead a saved file or rubric-scored answer, please identify that scope before expansion.

Codex implemented and verified these tasks on behalf of Ziao Yang, with disclosed assistance. Only public NOAA/USGS workflows are included; no conversation records or private data are shipped. CHANGELOG includes an Unreleased entry.