[CIR] Enable cir.bool binops

[andykaylor]

Add !cir.bool to the list of types accepted by the binop rewriter. Although we don't generate boolean binops from logical operations on C++ boolean values, there will be cases where such operations are useful while generating conditions for other operations, such as the overflow checks needed for array new handling.

[bcardosolopes]

Nothing wrong with using plain CIR checks, but if these pieces change we'll have to rewrite this testcase. Either (a) put the original C source in a comment or (b) write the test C -> LLVM IR directly (btw, we usually avoid writing LLVM IR dialect checks because for parity with LLVM IR, it's easier to compare the final IR with OG).

[andykaylor]

Nothing wrong with using plain CIR checks, but if these pieces change we'll have to rewrite this testcase. Either (a) put the original C source in a comment or (b) write the test C -> LLVM IR directly (btw, we usually avoid writing LLVM IR dialect checks because for parity with LLVM IR, it's easier to compare the final IR with OG).

I'm not sure it's possible to generate these cases from C/C++. If I have code like this:

bool f(bool a, bool b) {
  return a || b;
}

we generate a bunch of cir.ternary (presumably because the semantics for the || operator mean that the rhs expression might not be evaluated. That seems like something we may want to reevaluate at some point, but that the way it is now.

The reason I put up this PR is that when I implemented the array new support for the case with a variable array size, I needed to generate a binop or for the overflow handling, but the assert I'm modifying here rejected it during lowering. I threw in and and xor here just because they seemed logical to test.

[bcardosolopes]

Gotcha, thanks for the further explanation, LGTM