Sema: fix OOB access in coerceTupleToStruct

[wrongnull]

Closes #19529

[wrongnull]

not sure if I should add test for this

[Rexicon226]

Probably should add a test yes :)

[jacobly0]

not sure if I should add test for this

The default answer to this question is always yes.

[jacobly0]

Suggested change

	.struct_type => if (ip.loadStructType(inst_ty.toIntern()).field_names.len > 0)
	ip.loadStructType(inst_ty.toIntern()).field_names.get(ip)[tuple_field_index]
	else
	try ip.getOrPutStringFmt(sema.gpa, "{d}", .{tuple_field_index}, .no_embedded_nulls),
	.struct_type => ip.loadStructType(inst_ty.toIntern()).fieldName(ip, tuple_field_index).unwrap() orelse
	try ip.getOrPutStringFmt(sema.gpa, "{d}", .{tuple_field_index}, .no_embedded_nulls),

Note that the same thing can be done in the .anon_struct_type case.

[jacobly0]

Although, really it should probably be more like:

        const maybe_field_name = switch (ip.indexToKey(inst_ty.toIntern())) {
            .anon_struct_type => |anon_struct_type| anon_struct_type.fieldName(ip, tuple_field_index),
            .struct_type => ip.loadStructType(inst_ty.toIntern()).fieldName(ip, tuple_field_index),
            else => unreachable,
        };
        const struct_field_index = if (maybe_field_name.unwrap()) |field_name|
            try sema.structFieldIndex(block, struct_ty, field_name, field_src)
        else
            tuple_field_index;

[wrongnull]

done

[jacobly0]

I was trying to avoid the extra call to getOrPutStringFmt in the .anon_struct_type case, but the new code is certainly reasonable enough that I won't hold up the merge over it.

[wrongnull]