Fix :calendar crash on OTP 29 for IANA "24:00" transition times (fixes #166)

[patrols]

Fixes #166.

Problem

Erlang/OTP 29 tightened :calendar.time_to_seconds/1 to reject hours outside 0..23. The IANA tz database legally uses "24:00" for end-of-day transition boundaries, which Tzdata.Util.time_for_rule/2 parses into an hour-24 datetime (e.g. {{Y, M, D}, {24, 0, 0}}).

Tzdata.PeriodBuilder.datetime_to_utc/3 and Tzdata.Util.datetime_to_utc/3 pass that datetime to :calendar.datetime_to_gregorian_seconds/1 (which calls time_to_seconds/1 internally), so on OTP 29 the :tzdata_release_updater crashes during period building (as reported in #166):

** (FunctionClauseError) no function clause matching in :calendar.time_to_seconds/1
   (stdlib 8.0) calendar.erl:873: :calendar.time_to_seconds({24, 0, 0})
   (stdlib 8.0) calendar.erl:275: :calendar.datetime_to_gregorian_seconds/1
   (tzdata 1.1.3) lib/tzdata/period_builder.ex:381: Tzdata.PeriodBuilder.datetime_to_utc/3

OTP ≤ 28 accepted {24, 0, 0} and returned the value for the equivalent next-day 00:00, so this only surfaces after upgrading to OTP 29.

Fix

Add Tzdata.Util.datetime_to_gregorian_seconds/1, which computes the value directly:

:calendar.date_to_gregorian_days(date) * 86_400 + hour * 3600 + minute * 60 + second

This handles any hour (including 24) and is numerically identical to :calendar.datetime_to_gregorian_seconds/1 for hours 0–23 — 24:00:00 on day D ≡ 00:00:00 on day D+1. Both datetime_to_utc/3 implementations route through it.

Tests

• Added a regression test in test/tz_util_test.exs covering :utc / :standard / :wall modifiers with a 24:00 datetime.
• Full suite green on OTP 29.0.1 / Elixir 1.20.0 and OTP 28.5 / Elixir 1.19.5.

🤖 Generated with Claude Code