feat: add temp eclipse adjustment logic

[braddf]

Pull Request

Description

Apply a manual multiplicative adjustment to national solar forecasts for the partial solar eclipse on the evening of 12 August 2026 (~91% UK, ~88% NL, roughly 17:30–19:30 UTC).

ECMWF include the eclipse in their irradiance, but PVNet has never seen one in training and ignores it — a test run showed no kink over the eclipse window — so every model we serve over-forecasts through it. Same problem and same fix as 29 March 2025 (openclimatefix-archives/uk-pv-national-gsp-api#404), extended to NL and v1.

Multipliers come from James' and Sukh's tables (GB half-hourly, NL quarter-hourly) and live in one new module, internal/eclipse.py.

Scope

• National only — GSP and province forecasts are untouched, as in 2025.
• Forecasts only — PV Live / NED NL report the real dip, so generation is served as-is.
• All natonal models, since none of them reflect the eclipse.
• v0 (/national/forecast, /gsp/0/forecast, live paths of /gsp/forecast/all/) and v1 (/regions/{id}/forecast, /forecasts/snapshot) for GB and NL.

/forecasts/period and the pre-warmed caches need no changes: the period endpoint rejects region_type=national, warm_all_v1_caches skips NATION, and the v0 /forecast/all/ warm already excludes gsp 0.

Testing on an ordinary day: ECLIPSE_DATE=2026-08-10 uv run quartz-api moves the tables to that date. It's a plain env var read with a hardcoded default — no config or Terraform changes — so leave it unset in deployed environments.

Temporary. Once the stored values are corrected after the event, set ECLIPSE_ENABLED = False, then delete the module and its call sites. Tracked in openclimatefix/client-private#581.

Checklist:

• My code follows OCF's coding style guidelines
• I have performed a self-review of my own code
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works
• I have checked my code and corrected any misspellings