Fix Penman-Monteith daily solar radiation input (#152)

[kmich]

Summary

Fixes #152.

et0_penman_monteith() expects solar radiation as a 24-hour daily mean in W/m², but the coordinator was passing the weather station's instantaneous reading. Because the algorithm converts that value to MJ/m²/day, a noon reading could be treated as though it applied for 24 hours and substantially inflate ET₀.

Changes

• Snapshot the completed day's accumulated solar irradiation (Wh/m²) at the local day boundary.
• Persist the previous completed day's irradiation across Home Assistant restarts.
• Promote a saved "today" accumulation to "previous day" when HA restarts after midnight.
• Feed Penman-Monteith previous_day_whm2 / 24 instead of instantaneous solar W/m².
• Leave the PM sensor unavailable until a complete measured solar day exists, rather than publish a knowingly invalid estimate.
• Update the PM algorithm test to use a realistic daily-mean solar value.
• Add persistence coverage for the completed-day solar promotion path.
• Document the correction in CHANGELOG.

This intentionally takes the strict FAO-56 path rather than extrapolating a partial day's solar curve, avoiding unstable dawn estimates and errors after downtime/restarts.