Zoned format ffi

[sffc]

#5940

[sffc]

Discussed with @Manishearth and it seems the simplest solution is to return an error from this function if the option is not set.

Other solutions considered and eliminated:

1. Have multiple time zone types: not great because it requires multiple formatter types
2. Make zone variant non-optional: not great because you only need it for specific non-location formatting

[sffc]

@robertbastian says this should be an error and not try to do anything clever

[sffc]

Finally, the NeoZonedDateTimeFormatter FFI is producing correct output. The amount of code in FFI instead of Rust is a little bit more than I'd like (a few lines of business logic that aren't just argument marshaling), but it's fine. We could move some of it into Rust in the future by adding a method to load "all data required for GenericLong formatting", for example, so that we don't need to figure out which load functions to use in each constructor.

I'm still a bit stuck on what to do in the format_iso function. Since we have only one NeoZonedDateTimeFormatter type, we cast all of the formatters into the most general one, which requires TimeZoneInfo<Full>, even if the zone style is more specific than that. This is incompatible with the notion that we return an error if the FFI TimeZoneInfo fields were not populated when we want them populated.

I'm trying to think of solutions. One is that I could make a custom input type specifically for FFI (one that implements all required icu_datetime::scaffold::GetField and, by necessity, icu_datetime::scaffold::UnstableSealed) which sets an error if a field is accessed but is in its None state. This is not pretty but I think it gets the job done? EDIT: Actually, no, it doesn't quite work because the fields are loaded eagerly.