fix: model Uri userinfo username/password as nullable

[OmarAlJarrah]

Summary

• ParsedComponents stored username/password as non-null strings defaulting to "", which collapsed "absent" and "present-but-empty" into the same value before serialization ever ran.
• That broke the Uri PRESERVE profile's own documented equality contract: http://@h/ re-serialized to http://h/ (losing the @), and http://u:@h/ re-serialized to http://u@h/ (losing the trailing :), so two syntactically distinct URIs compared equal.
• username/password are now modeled as String? through the parse pipeline: null means the component wasn't present at all, "" means it was present but empty. Uri's serializer (and therefore equals/hashCode, both derived from uriString), userInfo accessor, structured resolver, and normalizer all now preserve that distinction, and Uri.Builder round-trips it via newBuilder().build().
• The Url (WHATWG) profile has no absent-vs-present-empty userinfo distinction in its own spec, so its parser, serializer, and public username/password accessors are unchanged in behavior (they still treat "" as absent).

Test plan

• ./gradlew :kuri:jvmTest — new regression tests added to UriTest, UriParserTest, UriBuilderTest, UriBuilderUserInfoTest, and an existing ResolverTest case updated for the corrected behavior
• ./gradlew :kuri:ktlintCheck
• ./gradlew :kuri:detekt
• ./gradlew :kuri:jvmApiCheck and ./gradlew :kuri:klibApiCheck — no public API shape change (ParsedComponents is internal), confirmed with no diff; androidApiCheck could not run in this environment (no Android SDK configured), which is a pre-existing environment gap unrelated to this change
• ./gradlew :kuri:jsNodeTest (commonMain touched)
• ./gradlew :kuri:macosArm64Test (native compile sanity check)
• ./gradlew :kuri:koverVerifyJvm

Closes #104