Go's url.Parse rejects bare '%' and other invalid percent-encoding
sequences, returning an error which causes obfuscateURLString to
return "?". The url crate silently re-encodes them as '%25', so
add an explicit pre-check matching Go's behavior.

Fixes fuzzing testcase: http_fuzzing_594901251

Go's url.Parse stores "." and ".." path segments literally, while the
url crate's join() resolves them via RFC 3986 normalization (making them
empty after stripping the base). Return the original input when
go_like_reference returns empty for a non-empty input that already
passed all error checks (control chars, invalid percent-encoding).

Fixes fuzzing testcase: http_fuzzing_3638045804

Go's url.Parse succeeds for relative URLs (like "0") and applies
path-digit removal to them. The Rust code was returning early from
the go_like_reference path without applying digit removal.

Add remove_relative_path_digits() helper and call it for relative URL
results when remove_path_digits=true.

Fixes fuzzing testcase: http_fuzzing_1928485962

…URLs

Go's url.shouldEscape for encodePath does not allow !, ', (, ), * even though
RFC 3986 considers them valid sub-delimiters in path segments. The url crate
follows RFC 3986 and keeps them unencoded. Post-process go_like_reference
output to encode these characters to match Go's behavior.

Fixes fuzzing testcase: http_fuzzing_4273565798

Go's validEncoded() has an explicit allowlist for !, ', (, ), * so these
are only re-encoded when the path has non-ASCII chars (which forces Go to
call escape() instead of using RawPath). For pure-ASCII inputs, Go's
EscapedPath() returns the RawPath unchanged, keeping ! as-is.

Only apply encode_go_path_chars() when the original input contains non-ASCII.

Fixes fuzzing testcase: http_fuzzing_1457007156

Go's url.Parse percent-encodes non-ASCII chars in fragments (e.g., '#ჸ' →
'#%E1%83%B8'). Our early-return fragment handler was returning the raw
fragment without encoding. Delegate non-empty fragments to go_like_reference
which uses the url crate's join() to correctly encode them.

Fixes fuzzing testcase: http_fuzzing_1092426409