Use a 365.25 denominator for the JULIAN day count

[dylanpulver]

DayCountTypes.JULIAN documents itself at day_count.py:60 as "actual divided by 365.25 as in Julian Calendar", but year_frac divides by 365 at day_count.py:284. A Julian year is 365.25 days by definition, so the member as shipped returns exactly the same accrual factor as ACT_365F and carries no distinct meaning.

On current master, 1 Jan 2019 -> 21 May 2019:

ACT_365F   acc_factor=0.3835616438356164 den=365
JULIAN     acc_factor=0.3835616438356164 den=365.0   <- identical

after: JULIAN acc_factor=0.38329911019849416 den=365.25.

How this happened (310bab06, "Significant edits to improve DiscountCurve..."): that commit renamed SIMPLE -> JULIAN, rewrote the comment to say 365.25, and replaced den = G_DAYS_IN_YEAR with a hardcoded den = 365.0 — the old constant's value rather than the new definition's. It also repointed the test at DayCountTypes.ASTRONOMICAL, which does not exist; 3c1c7f0c repaired the test to JULIAN two days later but left the pre-rename expected value, so the test has been pinning 365 ever since. G_DAYS_IN_YEAR is still imported at day_count.py:12 and is now otherwise unused, which is the leftover of that inline.

Change: den = 365.25, and the unit test's expected value re-derived from the definition rather than from the implementation — 140 days / 365.25 = 0.38330 — plus an assertion that JULIAN does not collapse onto ACT_365F.

Measured, same command and venv (Python 3.13, pytest unit_tests, which is what testpaths gives CI):

• before, unmodified master 8ba5e2c0: 957 passed
• after: 957 passed

Two mutants, to show the test is not vacuous:

• revert day_count.py only, keep the new test -> assert 0.3836 == 0.3833, 1 failed / 9 passed
• naive fix den = G_DAYS_IN_YEAR (restoring the now-dead import) -> same failure, which is the point: 365 in any spelling is not the Julian year

How I found it: not from real FinancePy usage. I was systematically reading day-count convention tables across quant libraries looking for branches that are byte-identical to a sibling while their own comment says otherwise. JULIAN matched, and the git history confirmed the intent rather than my guess at it.

Not tested: regression_tests/ (not in testpaths) — but grep -rn JULIAN finds no use of this member anywhere in financepy/ outside its own definition, so no internal pricing path changes. If you intended JULIAN to stay a 365-denominator alias, the right fix is the comment instead and I'm happy to flip it.

This was written with AI assistance (Claude).