fix: preserve NL day-count additivity at leap-day boundaries

[kadyrbekovhamit-cyber]

Summary

• treat February 29 at the start of an NL interval consistently with the existing (d1, d2] convention
• prevent negative day counts for an empty leap-day interval
• add boundary and additivity coverage for both NL/365 and NL/360

Problem

leap_days_between excluded February 29 from the end boundary, as established by the existing test_example_2::case_4, but did not exclude it from the start boundary. Therefore:

NL/365(2020-02-29, 2020-02-29) = -1 day
NL/365(2020-02-28, 2020-03-01) !=
  NL/365(2020-02-28, 2020-02-29) + NL/365(2020-02-29, 2020-03-01)

The change includes February 29 in the start-side adjustment (ordinal() >= 60) while deliberately retaining the current end-side < 60 comparison. This preserves the library's existing (d1, d2] financial interval convention.

Validation

• cargo test --release --features nonumpy: 227 passed, 10 ignored
• cargo fmt --check
• mutation check: restoring the old start-boundary predicate makes the new regression test fail