opti: avoid set shadowstack for (call local.get) pattern

[HerrCai0907]

optimize shadowstack
For (call local.get) pattern, local variable won't be changed in called function. So it is not needed to store the value of local.get in shadowstack

• I've read the contributing guidelines
• I've added my name and email to the NOTICE file

[dcodeIO]

Apart from the comment below, I wonder if there is unsafe code in stdlib that assumes call operands to go to the stack. For instance

let foo = changetype<usize>(new Foo());
...
someFunc(changetype<Foo>(foo));

would be such a pattern, where the local foo is not placed on the shadow stack due to being typed as usize, and with the change the call wouldn't place the value on the shadow stack either, then risking that the value is collected too early. Not sure if this is actually done, but I guess we should reasonably check that this pattern isn't used?

[dcodeIO]

Instead of duplicating the function, needsSlot could take a third argument isCallOperand: bool = false perhaps, then returning !isCallOperand on local gets. Wdyt?

[HerrCai0907]

I don't find this pattern is std.

And I think it will be danger only in this scenario:

class A {}

function f3(a: A): void {
  __unpin(changetype<usize>(a))
  // there 'a' is unsafe
}
export function f2callback(v: usize): void {
  f3(changetype<A>(v));
}

declare function f2(v: usize): void; // setTimeout(()=> f2callback, 100)

export function f1(): void {
  let a = new A();
  __pin(changetype<usize>(a));
  f2(changetype<usize>(a));
}

Wdyt? Maybe this optimization is not work.