Introduce IR for execution

[fengb]

There's some direct instructions that require branches inside the body:

fn @"0x10 call"(id: usize) !void {
    const func = self.instance.funcs[func_id];
    switch (func.kind) {
        .imported => {},
        .implemented => {},
    }
}

This branch is actually known at module processing time, but currently there's no way of differentiating since the execution engine maps directly to the bytecode.

We should insert a custom IR so we can optimize the instructions before execution.