OptimizeInstructions and i64

[dcodeIO]

It appears that the OptimizeInstructions pass does not yet hand-optimize i64 instructions as much as it optimizes i32 instructions. This recently surfaced when we changed compilation of !someI64 to emit eqz(isTrueish) instead of using a dedicated isFalseish helper, leading to missed optimizations like

i64.const 0
i64.ne
i32.eqz

becoming

i64.const 0
i64.eq

but not being optimized further to

i64.eqz

initially leading to the assumption that this is caused by the expression collapsing very late, thus requiring --converge, which is not the case at it seems. Pinning it here so we don't forget :)