Fixes #654.

Supersedes #1557.

Problem

For an uncached DAILY rule, between() starts iteration at DTSTART and discards every occurrence through the lower bound. The example in #654 therefore replays roughly twenty years of daily occurrences to return one month.

Implementation

• Let uncached between() queries request an iterator positioned near the lower bound.
• For DAILY rules without COUNT, seek to the preceding aligned interval instead of replaying from DTSTART.
• Preserve the existing path for cached rules, subclasses, incompatible timezone-awareness, floating timezone objects, datetime boundaries, and unsupported recurrence shapes.
• Cover distant starts, multi-day intervals, BYHOUR, timezone conversion, datetime limits, cache population, and subclass iteration.

Performance

Measured against current master on an AMD Ryzen 7 8845H under WSL2 with CPython 3.13.15. Each result is the median of 11 CPU-pinned timeit batches after 10 warm-up calls, using the reproducer from #654.

This is approximately a 164x speedup for #654's 20-year, uncached DAILY workload. Both versions return the same 30 occurrences; reaching the query interval requires 7,337 iterator pulls on master and 33 with this change.

The benefit scales with the distance from DTSTART: a query starting roughly one month after DTSTART improved from 76.5 us to 45.7 us (1.68x). A query that cannot use the seek path remained effectively unchanged (38.05 us on master and 38.20 us with this change). These measurements are specific to the supported DAILY seek path, not a general rrule-wide speedup claim.

CI compatibility

The branch keeps two current test-suite compatibility fixes as separate commits:

• normalize the timezone property test to its UTC-aware Hypothesis input domain;
• materialize isoparser parameters as required by pytest 9.1.

Neither compatibility commit changes dateutil runtime behavior.

Validation

• The rrule.between is quite slow #654 reproducer returns identical results before and after the change.
• A 384-case differential matrix covering intervals, start hours, BYHOUR values, query distances, and inclusive/exclusive bounds had zero mismatches against the legacy path.
• Key rrule, timezone-property, and isoparser tests passed on CPython 2.7, 3.7, 3.12 with pytest 9.1, and 3.14.
• Python 3.13 install-state tox suite: 2041 passed, 47 skipped, 17 xfailed.
• Latest IANA tz master suite: 2040 passed, 47 skipped, 17 xfailed.
• Coverage: 88%.
• Docs, linkcheck, pre-commit, Darker, wheel, and sdist checks passed.

Disclosure

This change was developed with Codex assistance.