Fix byte-string URI userinfo encoding in Neo4j::Driver HTTP adapter

[fglock]

Summary

PerlOnJava incorrectly UTF-8-encodes a Perl byte string when serializing HTTP Basic Authentication userinfo in Neo4j::Driver::Net::HTTP::LWP. This causes the upstream Neo4j-Driver test suite to fail on both the JVM and interpreter backends.

CPAN impact

• Distribution: Neo4j-Driver 0.52
• Affected module selected by the tester: Neo4j::Driver::ResultColumns
• Failing test program: t/net-lwp.t
• Failing subtest: auth variations, test 4, uri with latin1 userid
• CPAN run: 20260908-215506-1375
• PerlOnJava result: 1 of 119 subtests failed; 1 of 21 test programs failed

The other failures reported by neither the suite nor the environment are absent: the remaining tests pass or skip normally because no Neo4j server is available.

Observed behavior

The upstream test passes a byte string containing the octets C4 80 as the Basic Auth principal and expects those octets to be percent-escaped directly in the URI:

expected: %C4%80
actual:   %C3%84%C2%80

The actual value treats the two Latin-1 bytes as Unicode characters U+00C4 and U+0080, then encodes each character as UTF-8 before percent escaping. The resulting URI is:

http://%C3%84%C2%80:@net.test/

The expected URI userinfo is:

http://%C4%80:@net.test/

The same test also verifies a genuinely Unicode password (U+0100); that case passes, so the failure specifically distinguishes byte-string and decoded-Unicode semantics.

Oracle and backend comparison

The unchanged upstream t/net-lwp.t passes completely under system Perl:

Files=1, Tests=14
Result: PASS

The focused test fails identically on both PerlOnJava execution backends:

• JVM backend: fails auth variations test 4
• Interpreter backend: fails auth variations test 4

This is therefore a shared PerlOnJava scalar/Unicode representation issue, not a compiler-backend difference, missing Neo4j server, timeout, or native dependency problem.

Suspected ownership

Neo4j::Driver::Net::HTTP::LWP deliberately handles the distinction as follows:

1. It checks utf8::is_utf8 on each authentication component.
2. It calls utf8::encode only for values marked as decoded Unicode.
3. It calls URI::Escape::uri_escape to percent-escape the resulting bytes.
4. It stores the escaped userinfo in a cloned URI.

System Perl preserves the unflagged C4 80 byte string through this path. PerlOnJava instead causes the bytes to be observed/serialized as characters before URI userinfo serialization, producing the extra UTF-8 expansion. The likely ownership is in PerlOnJava byte-string flag/representation handling or the bundled URI userinfo serialization path. The upstream adapter code should not be changed to special-case this input because its current logic matches standard Perl behavior.

Reproduction

Run the upstream test from the Neo4j-Driver 0.52 distribution:

t/net-lwp.t

The failure is:

Failed test 'uri with latin1 userid'
'http://%C3%84%C2%80:@net.test/' doesn't match '(?^i://%C4%80:@)'

Acceptance criteria

• Add a project-owned regression test for URI userinfo serialization of an unflagged byte string containing octets above 0x7f.
• Validate the regression test against system Perl before using it as the PerlOnJava oracle.
• Preserve direct byte escaping: C4 80 must become %C4%80, not UTF-8-expanded %C3%84%C2%80.
• Preserve decoded-Unicode behavior: U+0100 must continue to serialize as %C4%80.
• The focused regression passes on both JVM and interpreter backends.
• Neo4j-Driver 0.52 t/net-lwp.t passes completely on both backends.
• Avoid regressing ordinary ASCII userinfo, already percent-escaped userinfo, and URI behavior for normal decoded Unicode values.

Related byte/Unicode issues #1182, #1183, and #1189 may involve neighboring representation boundaries, but none currently covers URI userinfo percent-encoding or this reproducible failure.