Issue Description:

let nsurl = URL(string: "h:#ash")!
assert(nsurl.absoluteString == "h:%23ash")

Here, the scheme is "h" and the path is "%23ash". Most other parsers would say that the scheme is "h", the path is empty, and the fragment is "ash".

Compare:

let nsurl = URL(string: "h:?ash")!
assert(nsurl.absolueString == "h:?ash")

The documentation says URL implements RFC-1808, but this would seem to be a point of non-compliance. From the RFC:

URL         = ( absoluteURL | relativeURL ) [ "#" fragment ]

absoluteURL = generic-RL | ( scheme ":" *( uchar | reserved ) )   

generic-RL  = scheme ":" relativeURL   

relativeURL = net_path | abs_path | rel_path   

net_path    = "//" net_loc [ abs_path ]   
abs_path    = "/"  rel_path
rel_path    = [ path ] [ ";" params ] [ "?" query ]

So looking at the string "h:#ash"

• The URL rule says it must start with an absoluteURL or relativeURL

• The absoluteURL rules says it may start with generic-RL

• The generic-RL rule says that we split the string at the first ":", so scheme is "h" and relativeURL is "#ash"

• relativeURL may be a net_path (starts with "//"), abs_path ("/") or release_path

• rel_path says everything is optional; it could be an empty string.

• Therefore, considering the rule in URL which splits at the first "#", the fragment should be "ash"

In fact, the RFC actually recommends the fragment be split before doing any other parsing:

If the parse string contains a crosshatch "#" character, then the
   substring after the first (left-most) crosshatch "#" and up to the
   end of the parse string is the <fragment> identifier.  If the
   crosshatch is the last character, or no crosshatch is present, then
   the fragment identifier is empty.  The matched substring, including
   the crosshatch character, is removed from the parse string before
   continuing.