Url::port_int implementation

[daxpedda]

Some questions arose up while I was doing this implementation.

• I implemented this by adding a field port_int to Url, my initial implementation actually handled it by storing the actual port always in Url::port and when using the method Url::port() it would filter based on scheme defaults. In addition the Url::serialization would always contain the correctly removed default port. This implementation caused a lot of edge cases to arise, so I avoided it by doing it in a separate field.
• Currently when changing a scheme with Url::set_scheme, Url::port_int is removed if it was the scheme default, see Changing the scheme can break things #61. Probably it would be more intuitive that if Url::port_int is empty, ergo the user didn't define a port (default or not), then having the new port be the default port for that scheme makes sense. But if the user deliberately defined the default port on the URL, changing the scheme shouldn't change the port.

  // this makes sense
  let mut url = Url::parse("http://example.net").unwrap();
  url.set_scheme("https").unwrap();
  assert_eq!(url.port(), None);
  assert_eq!(url.port_int(), None);
  assert_eq!(url.port_or_known_default(), Some(443));
  
  // this is not intuitive
  let mut url = Url::parse("http://example.net:80").unwrap();
  assert_eq!(url.port(), None);
  assert_eq!(url.port_int(), Some(80));
  assert_eq!(url.port_or_known_default(), Some(80));
  url.set_scheme("https").unwrap();
  assert_eq!(url.port(), None);
  assert_eq!(url.port_int(), None); // why does this vanish, probably expected to stay 80
  assert_eq!(url.port_or_known_default(), Some(443)); // probably expected to stay 80
  
  // even weirder, but you shouldn't use a assigned port to a custom protocol anyway
  let mut url = Url::parse("custom-protocol://example.net:80").unwrap();
  assert_eq!(url.port(), Some(80));
  assert_eq!(url.port_int(), Some(80));
  assert_eq!(url.port_or_known_default(), Some(80));
  url.set_scheme("http").unwrap();
  assert_eq!(url.port(), None);
  assert_eq!(url.port_int(), None);
  assert_eq!(url.port_or_known_default(), Some(80));
  url.set_scheme("https").unwrap();
  assert_eq!(url.port(), None);
  assert_eq!(url.port_int(), None);
  assert_eq!(url.port_or_known_default(), Some(443)); // port changed completely, I guess this is potentially an issue with the current implementation too

  I don't think this is a big deal, I would argue that this is a pretty bad use-case the url crate might not want to support.
• Comparing two Urls with different Url::port_int might turn out to be equal, because it compares Url::serialization, and default ports for scheme aren't reflected in the serialization.

---

Personally I'm fine with last two points, I consider them edge-cases that are fine for the url crate to ignore.

Fixes #706.

[daxpedda]

Potentially I could make an enum like this:

enum Port {
    None,
    Default(u16),
    Custom(u16),
}

[daxpedda]

If use the enum from above I could potentially pass it down from here, reducing the complexity.

[daxpedda]

This might be considered a breaking change? I consider checking for user input regardless of defaults "more" correct, this might have been considered a bug until now.
This can easily be removed.