fix: derive Jalali conversions from the astronomical calendar model

[majiidd]

Problem

Reported: persiantools converts Gregorian 623-01-01 to Jalali 1-10-11 while astronomical converters such as time.ir give 1-10-10, and the kabiseh (leap year) logic was internally inconsistent.

Investigation confirmed two real defects on master:

• Conversion bug (Gregorian ≤ 1600): the year <= 1600 epoch branch in to_jalali was arithmetically wrong — 185,426 days in 622–2130 disagreed with the model the rest of the code implements (e.g. 623-03-21 → 2-1-1, skipping Esfand 30 of leap year 1), and 237 days raised ValueError.
• Kabiseh inconsistency: is_leap/days_in_month used the ICU4X NON_LEAP_CORRECTION_SET while to_jalali/to_gregorian used plain 33-year arithmetic — so 2124-03-20 was unrepresentable (ValueError) and 1503-12-30 / 1504-01-01 both mapped to 2125-03-21.

What changed

All conversions, ordinals, and leap checks now derive from a single internal model (_days_before_year + is_leap):

• Epoch corrected to the documented Solar Hijri epoch: Friday 1 Farvardin 1 = 22 March 622 CE proleptic Gregorian (19 March 622 Julian, R.D. 226896).
• Years 1–1177 follow the astronomical Persian calendar (vernal equinox at 52.5°E, Calendrical Calculations), generated with roozbehp/persiancalendar — the model that reproduces the official leap-year table of the Iranian calendar authority (Calendar Center, Institute of Geophysics, University of Tehran, 1206–1498) exactly. Encoded as 133 flip years against the 33-year rule.
• Five borderline flip pairs overridden — (978, 979), (1011, 1012), (1044, 1045), (1077, 1078), (1176, 1177): their effects would reach into Gregorian 1601+, where all established implementations agree with the 33-year value, and each hinges on the equinox missing the midday cutoff by 36 s–11 min (inside ephemeris model uncertainty).
• Years 1178–2987 unchanged (33-year rule + ICU4X corrections); 2988+ unchanged (plain 33-year rule).

Behavior changes

Range	Change
Gregorian 622-03-22 … 1568-03-20	Corrected to the astronomical model (623-01-01 → 1-10-10; 237 former ValueError days fixed)
Gregorian 622-03-21	Was 1-1-1; now pre-epoch → ValueError
is_leap, years ≤ 1177	Changes for 133 flip years (e.g. is_leap(1): True → False)
Gregorian 1568-03-21 … 2124-03-19	No change (verified day-by-day)
Gregorian ≥ 2124-03-20	Conversions now follow the correction set (Norouz 1503 = 2124-03-20; no more ValueError/collisions)
toordinal() / fromordinal()	All ordinals shift by −1 (epoch moved one day later); round-trips preserved; may affect callers who persisted raw ordinals

Validation

• Compatibility sweep: all 191,100 days in 1601-01-01 … 2124-03-19 convert identically to master.
• Astronomical reference: Norouz and leap flag match roozbehp/persiancalendar for every year 1–3000 except the five documented overrides.
• JDF reference: all 202,709 days of Persian years 947–1501 identical to the jdf.scr.ir algorithm.
• Integrity: all 3,424,878 representable days (622-03-22 … 9999-03-20) round-trip exactly with monotonic ordinals; every year length matches is_leap.
• New tests: epoch (Friday 622-03-22), the 623-01-01 regression, a 32-row ancient Norouz table, ancient leap flips, plus the official 1206–1498 kabiseh table test (tests/test_official_kabise.py, all 293 years). make check green; opt-in performance tests pass.

Note: the jdf.scr.ir algorithm itself also yields 1-10-11 for 623-01-01 — the reported time.ir value comes from the astronomical model, which this PR adopts for ancient years on the strength of the documented epoch (Persian Wikipedia: جمعه ۱ فروردین سال ۱ = ۲۲ مارس ۶۲۲) and the official authority's astronomical practice.

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

❌ Patch coverage is 96.68050% with 8 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.19%. Comparing base (9b95cee) to head (7899a38).

Files with missing lines	Patch %	Lines
persiantools/jdatetime.py	96.17%	8 Missing ⚠️
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #69      +/-   ##
==========================================
- Coverage   98.47%   98.19%   -0.29%     
==========================================
  Files           5        5              
  Lines        1117     1220     +103     
==========================================
+ Hits         1100     1198      +98     
- Misses         17       22       +5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.