Use empty blocks instead of nops for empty scopes in IRBuilder

[tlively]

When IRBuilder builds an empty non-block scope such as a function body,
an if arm, a try block, etc, it needs to produce some expression to
represent the empty contents. Previously it produced a nop, but change
it to produce an empty block instead. The binary writer and printer have
special logic to elide empty blocks, so this produces smaller output.

Update J2CLOpts to recognize functions containing empty blocks as
trivial to avoid regressing one of its tests.

[kripken]

These nested locations surprise me a little. Was the old parser really emitting an empty block there? I am pretty sure it wasn't.

[tlively]

I think what's happening here is that the empty function gets parsed with an empty block and then DAE adds the local.set in front of the empty block rather than inside it. Since the empty block is no longer the function body, the printer happily starts printing it.

I do know that when we switched over to the IRBuilder-based text parser, we ended up adding a lot of nops to a lot of tests, but I don't think the old text parser could have been doing anything different in this case, since it would have had to do something or other for the empty function body.

[kripken]

I see. Seems reasonable.

Maybe worth checking if say the unoptimized code size tests in Emscripten change with this, just to be safe,

./test/runner other.test*unoptimized*

(optimized size should not be able to change as empty blocks get removed).

[tlively]

Hmm, don't those tests skip Binaryen entirely because they are run with -O0?

[tlively]

Also this only affects the text parser for now.

[kripken]

Oh right, nevermind my worries then.