Handle nonmappable code points in Document::encoding_parse_a_url (#37541)

This is a followup to https://github.com/servo/servo/pull/33825. Using
`Encoder::encode` introduced a subtle bug: That function will silently
replace nonmappable code points (such as `㐀` in euc-jp). The url spec
however expects nonmappable characters to be treated differently. There
is actually an open bug in the `rust-url` repo about this:
https://github.com/servo/rust-url/issues/649, with the conclusion
apparently being that this should not be implemented by the url crate
itself.

Gecko implementation of the equivalent algorithm for reference:
https://searchfox.org/mozilla-central/rev/d52edf7ea4236446e118a2edc815023c5479663f/netwerk/base/nsStandardURL.cpp#116-172.

Testing: More web platform tests pass

Part of https://github.com/servo/servo/issues/5601

---------

Signed-off-by: Simon Wülker <simon.wuelker@arcor.de>

components/script/dom/document.rs

@@ -3672,7 +3672,9 @@ impl Document {
         // Step 5. Return the result of applying the URL parser to url, with baseURL and encoding.
         url::Url::options()
             .base_url(Some(base_url.as_url()))
-            .encoding_override(Some(&|s| encoding.encode(s).0))
+            .encoding_override(Some(&|input| {
+                servo_url::encoding::encode_as_url_query_string(input, encoding)
+            }))
             .parse(url)
             .map(ServoUrl::from)
     }