Optimize load-extend sequence

[lexaknyazev]

These two functions are semantically the same but the first uses one more instruction:

export function foo(): u64 {
  const a = load<u8>(16);
  return <u64>a;
}

export function bar(): u64 {
  return <u64>load<u8>(16);
}

 (func $module/foo (result i64)
  i32.const 16
  i32.load8_u
  i64.extend_i32_u
 )
 (func $module/bar (result i64)
  i32.const 16
  i64.load8_u
 )

[MaxGraey]

Good catch! Will add this optimization rule for binaryen.
For bar it works due to it produced by frontend of compiler while foo required works for backend (binaryen)

[MaxGraey]

Hopefully it will soon be available in Binaryen: WebAssembly/binaryen#4307