Don't use static function argument for spawn

[rossberg]

The explainer suggests that spawn takes a function index. That might work for simple local parallelism, but would make it impossible to build any form of abstractions around threads and their management. I think this clearly needs to be a function reference to be sufficiently general.

[yamt]

a funcref is fine for me.
it implies to make it have a type imm as call_indirect does, right?

[abrown]

but would make it impossible to build any form of abstractions around threads and their management

@rossberg, I was going to switch the explainer to use funcref but then I started to wonder, "is this statement true?" Maybe I'm missing something but the parallel frameworks I'm familiar with would typically define their own thread entry point — a statically-known function — and then dispatch from there to the function (e.g., from a funcref table) doing the work, after they've done whatever framework logic is necessary. I'm open to switching this over but perhaps we just need to use the generality argument here?

[rossberg]

Well, you'll have to pass that static function another function representing the thread to create as a parameter somehow. The cheap and most natural to do that is as a reference. Yes, you could take the detour through a global table by stuffing all references you ever want to call it with into that, but that complicates the interface with extra indirection and global state, which is totally unnecessary, unless you absolutely have to cook up an integer for other reasons.