Fix GH-17317: ResourceBundle crash on undefined iterator key

[nielsdos]

The next() and rewind() calls should be the ones fetching the data. We skip holes in the resource bundle in the same way that holes in arrays are skipped.

[cmb69]

Thank you! I haven't checked very thoroughly, but this looks like a sensible solution. There is, however, an issue regarding the ::count() method:

var_dump((new ResourceBundle('', NULL))->get('calendar')->get('buddhist')->count());

reports 12, but traversal only shows 10 items. We can't change ::count() for BC reasons (would break "manual" for iterations using ::get()). If we want to avoid this dichotomy (at least for stable branches), we could return null for key and value for missing elements (I was missing that resourcebundle_iterator_current() can simply return NULL), but that still would leave a difference regarding ::getError*() (see middle part of #17318 (comment)).

Probably not a big deal, so I'm fine merging this PR as is into PHP-8.3.

[nielsdos]

There is, however, an issue regarding the ::count() method:

Indeed, this one is unfortunate but we should preserve BC.

we could return null for key and value for missing elements

Sounds like a typical PHP solution, and inconsistent with how arrays work imo.

I was missing that resourcebundle_iterator_current() can simply return NULL

You can only do this if the iterator is invalid, so it's not a valid alternative solution (if you're talking about returning a NULL pointer).

but that still would leave a difference regarding ::getError*() (see middle part of #17318 (comment)).

This is consistent with how arrays work:

<?php
$array = [0 => "a", 2 => "b"];
$array[1]; // warning & returns NULL
foreach ($array as $k=>$v) { var_dump($k, $v); } // skips 1