Fix GH-24350: Stale external tz database read after module restart

[lazerg]

PHP_MINIT_FUNCTION(date) registered its INI entries before resetting php_date_global_timezone_db. Registering date.timezone runs OnUpdate_date_timezone(), which checks the value against DATE_TIMEZONEDB. On an in-process engine restart (Apache graceful reload, FrankenPHP worker restart), that pointer still points into the PECL timezonedb .so that the previous shutdown unloaded, so startup reads unmapped memory and crashes in seek_to_tz_position(). The reset now happens before REGISTER_INI_ENTRIES().

I checked it with a small embed host on macOS: it loads timezonedb as a shared extension, calls php_embed_init()/php_embed_shutdown() twice, and maps a PROT_NONE region over the address range the unloaded .so used. Before the fix, the second startup crashes in seek_to_tz_position(). After the fix, it starts normally and still uses the external database. There is no test, because run-tests cannot restart the engine inside one process.

Fixes GH-24350

[lazerg]

The FreeBSD job fails because the FreeBSD 13 package mirror returns 404. It fails the same way on other PRs today, so my change does not cause it.

[ndossche]

Thanks

[lazerg]

The macOS job fails on ext/zip/tests/oo_torrentzip.phpt. The same test fails on other PRs today (for example #24365 and #24367), so my change does not cause it.