Add support for safe key constant value propagation / folding that preserves "key-ness" about tailslide

[SaladDais]

May be more complicated than I thought:

key glob_uuid = "<some_uuid>";
// float but with an "int-like" constant
float int_like = 1;
float int_like_2 = int_like;
list list_glob = [glob_uuid, int_like, int_like_2];

default
{
    state_entry()
    {
        key local_uuid = "<some_uuid>";
        llOwnerSay(
            // string-like under LSO
            (string)llGetListEntryType(list_glob, 0) +
            // int-like under LSO
            (string)llGetListEntryType(list_glob, 1) +
            // also int-like under LSO
            (string)llGetListEntryType(list_glob, 2) +
            // float-like under both LSO and mono
            (string)llGetListEntryType((list)int_like, 0) +
            // float-like under both LSO and mono
            (string)llGetListEntryType([int_like], 0) +
            // key-like under both LSO and mono
            (string)llGetListEntryType([glob_uuid], 0) +
            // string-like under LSO
            (string)llGetListEntryType((list)glob_uuid, 0) +
            // key-like under both LSO and mono
            (string)llGetListEntryType([local_uuid], 0) +
            // string-like under LSO.
            (string)llGetListEntryType((list)local_uuid, 0)
        );
    }
}

Prints "311224343" under LSO but "422224444" under Mono. I suspect key values are usually stored on the heap as an LST_STRING, and the "key-ness" is specifically added by the instruction that pushes the entry onto the list in the list expression case (which is typed and forces key.)

keys appears to always lose key-ness as do global list "expressions" (really simpleassignables under LL's compiler,) as do list casts.

There's some similar logic for floats with "int-like" constants under LSO, but "float-ness" is never lost through casting a global because it doesn't have an existing heap entry that can be used like keys do.

[SaladDais]

Mostly handled in the LSO branch. Currently the constant value propagation more closely matches Mono so using non-strict global validation or optimization on scripts meant to run on LSO will break them. Would need LSO-specific constant value propagation to support it, and I'm not really interested in doing that right now.