Investigate options of reducing the base overhead of `async` functions as well as overhead per `await`.

[mkustermann]

We pay a high code size cost for async functions atm and we should investigate possible improvements.

[mkustermann]

To determine the priority of this, I have (hopefully correctly) analyzed a version of a big internal app (GP) with the following results

type	count	wrapper	closure	total
async	3909	1349276	3826420	5175696
async*	11	4280	37652	41932
sync*	52	4776	5348	10124

• Around 4k functions out of 92k are async functions.
• The wrapping functions take up 1.35 MB, 6% of RX
• The inner closure functions (which includes application code plus await overhead) take up 3.8 MB, 17.5% of RX

RX makes up 65% of the final AOT app (app.so).

So I think we should prioritize this.

[mkustermann]

/cc @alexmarkov @mraleph