Add is_reference_type()

[adlerjohn]

Similar to #818, add a is_reference_type::<T>() function that is replaced by true or false at compile time depending on if the type is a reference type or not. This is needed so that the stdlib can do different run time logic depending on whether a type is a pointer to data vs data in a register.

Note that currently structs with a single single-word field are actually reference types, so the stdlib cannot use size_of::<T>() against the word size exclusively for such runtime checks.

[mohammadfawaz]

In the near future, we won't really know if a type is going to be implemented in registers or not. It is very reasonable for the IR to decide to unwrap a local struct of 2 ints into 2 two separate ints, or an array into its individual elements (if the array accesses have fixed indices), in which case the result of is_reference_type::<T>() becomes incorrect. Where a variable is implemented is really just an implementation detail. We can imagine a (legal but really horrible) compiler that implements everything in storage!

That being said, I may have to catch up on some Slack message to understand the context better. I imagine this has something to do with asm blocks.

[otrho]

Correct. This isn't something we want to keep once we add intrinsics and remove the ASM blocks from the stdlib. But until then we need something to help out.

[adlerjohn]

To add some context here for permanence: store is the most obvious example of this. It needs to know if something is a reference or not so that it can know its size, which it needs to know if it should use SWW or SWWQ.

Currently, store assumes all T are one-word primitives.

sway/sway-lib-std/src/storage.sw

Lines 3 to 7 in 4bbccc4

	pub fn store<T>(key: b256, value: T) {
	asm(r1: key, r2: value) {
	sww r1 r2;
	};
	}