perf(webidl): optimize createRecordConverter()

[AaronO]

Cuts self-time by ~6x, 172ns/iter => 22ns/iter benched on 1M Response builds / HeadersInit calls

This should be functionally identical to the old code since:

1. for...in loops on objects iterate over all enumberable string keys of an object (including inherited)
2. Then we filter out inherited keys with .hasOwnProperty(...)

Note: WebIDL / WPT is coupled to a subpar implementation so our hands are tied, unless an isProxy fast path check isn't too expensive

Benchmarks

No fast path:
resp_w_bh:           	n = 1000000, dt = 1.295s, r = 772201/s, t = 1294ns/op
Fast path with isProxy check:
resp_w_bh:           	n = 1000000, dt = 1.146s, r = 872600/s, t = 1146ns/op
Forced fast path:
resp_w_bh:           	n = 1000000, dt = 1.112s, r = 899281/s, t = 1112ns/op

Pending on #12288

[lucacasonato]

This is not equal. for .. in also enumerates prototype fields, while ReflectOwnKeys does not.

{
  const obj = {
    a: 1,
    get b() {
      return 2;
    },
  };
  Reflect.getPrototypeOf(obj).c = 3;

  for (const key in obj) {
    const val = obj[key];
    console.log("for .. in:", key, val);
  }

  // for .. in: a 1
  // for .. in: b 2
  // for .. in: c 3
}



{
  const obj = {
    a: 1,
    get b() {
      return 2;
    },
  };
  Reflect.getPrototypeOf(obj).c = 3;

  const keys = Reflect.ownKeys(obj);
  for (const key of keys) {
    const val = obj[key];
    console.log("Reflect.ownKeys:", key, val);
  }

  // Reflect.ownKeys: a 1
  // Reflect.ownKeys: b 2
}

[lucacasonato]

Actually, NVM. Inherited keys are filtered out via hasOwnProperty