tzdb.cpp: Fall back to numeric offset when the current code page cannot represent the time zone short ID

[cpplearner]

This PR adds a fallback time zone abbreviation of the form -05 or +0530 when the system's code page cannot represent the localized time zone short ID produced by ICU. Previously such time zone short ID would result in ERROR_NO_UNICODE_TRANSLATION exceptions.

The numeric offset form (-05 or +0530) is chosen because it is similar to the time zone abbreviation in tzdb, and distinguishable from ICU's output.

The behavior depends on the current user locale and the current code page. For ICU 77.1, the following locales have time zone short ID with non-ASCII characters, and thus might be affected by this fallback: am, ar, as, bg, blo, brx, ckb, cs, dz, fa, ff_Adlm, ga, gd, he, hi, ii, km, ks, ks_Deva, ku, mai, ml, mni, nqo, sa, sat, sd, sd_Deva, si, sk, vi, xnr. In addition, the following locales have non-ASCII character (U+2212 or U+2013) as the minus sign, and thus are also affected: et, eu, fa, fr, lij, lt, se, sv.

No tests are added because the en locale is unaffected.

Fixes #3158
Fixes #5482

[StephanTLavavej]

Thanks! 😻 I adjusted the code slightly, then manually verified that it still correctly prints both minutes and whole hours (by brain-surgery nulling out the abbrev pointer in the debugger).

[StephanTLavavej]

I'm mirroring this to the MSVC-internal repo - please notify me if any further changes are pushed.

[StephanTLavavej]

Thanks for figuring out how to fix this in a reasonable way! 😻 🗺️ 🕥

[TheStormN]

Sorry for commenting after a merge, but does this release() call means we are leaking memory?

[cpplearner]

We aren't leaking memory. The returned __std_tzdb_sys_info* will be stored in a unique_ptr<__std_tzdb_sys_info, _Tzdb_deleter<__std_tzdb_sys_info>>, which will call __std_tzdb_delete_sys_info in its destructor, which will perform delete[] _Info->_Abbrev;