Disable lazy selector registration check for OLDABI non-compat

[HendrikHuebner]

WebAssembly's linear memory does not guarantee that static/heap alocated selector strings are greater than a certain numeric value. Therefore, we can't reliably assume that sel->index < selector_list->size() means the selector is registered already. The pointer value might be smaller than the size of the selector_list.

This patch uses tsl::robin_map for WebAssembly instead. Perhaps this could be optimized in the future by extending the selector union objc_selector on WASM to have an additional entry to mark it as registered? I think this would also require a compiler change however, so this seems like the easiest option for now.

[davidchisnall]

Do we actually need the is-registered check with the v2 ABI? I think we now guarantee that all selectors are registered in __objc_load. The check is still there because someone might load something with the v1 ABI and do __objc_exec_class to load the classes but expect the selectors to be lazily resolved.

I think WebAssembly can just not expose the v1 entry point and skip the is-resolved checks. We can probably disable them entirely in any build configuration where we're not building v1 ABI support and stick in a #error if you ask for v1 ABI and WAsm in the same build?

[HendrikHuebner]

Alright, I #ifdef'd the check to always return true when the old ABI is not enabled.

I'll create a separate PR to update the build system to support a LIBOBJC_WASM mode, which disables the old ABI.
We'll also have to build libobjc2 as a static archive for WebAssembly.

[HendrikHuebner]

@davidchisnall please don't merge this yet - I think selLookup_locked has the same issue. I'll need to look into it first

[HendrikHuebner]

@davidchisnall The current #ifdef OLDABI version is wrong. __objc_load will the selector registration entirely because isSelRegistered returns YES all the time. That's why the Windows builds are failing. Should I revert this to the original version which uses the hash set for WASM?

[HendrikHuebner]

@davidchisnall I reverted it to the original patch. Can you restart the CI checks please?