Add br_if and br_switch.

[sunfishcode]

This adds br_if and br_switch to AstSemantics.md, essentially as suggested in #299 . The semantic description here is very brief; for full details see the corresponding spec PR, WebAssembly/spec#131 .

[rossberg]

Do we actually have consensus on br_switch? As far as I recall the discussion in #299, the only thing people could agree to (and not particularly enthusiastically) was adding break_if/break_continue. But br_switch is a significantly more complicated operator. It's fine changing the switch operator in certain ways (e.g., making it fallthrough by default, and/or making it table switch over a range), but I doubt we should have several switch operators.

In general, I'm still not sure where all this is going, and whether I agree with the direction.

[lukewagner]

@rossberg-chromium The argument that convinced me in #299 is that the natural way to transform a CFG into structured control flow is to generate tons of if-break's and switch-break's and so it makes sense to have these available as primitives (less ops to decode and no need to pattern-match to fold branches means faster decoding and better performance for dumb compilers). br_if and br_switch seem symmetric in this regard, as they are two pieces of the same algo; it'd be weird to only have one. In terms of engine complexity, I think it will be negligible (both switch and br_switch would decode into the same engine-internal IR op).

Also, with br_switch, you get fallthrough "for free", and so switch could keep its simple, high-level definition.

[sunfishcode]

After several offline discussions, I'm now working on a new hybrid proposal that attempts to merge the best of both approaches. I'll open a new PR when the new plan is ready.