lab-ops-hub: container validator's #577 backslash normalization overreaches into query and fragment — WHATWG keeps \ as data there, so ?q=a\b is fetched as ?q=a/b #620

Summary

The container validator's backslash normalization (#577, pkgs/lab-ops-hub/container/app.py in validate_crawl_url) replaces every \ in the whole input with / for http/https URLs. WHATWG treats a backslash as a path delimiter only in the scheme/authority/path portion of a special-scheme URL — in the query and fragment it is ordinary data and is kept verbatim. The container therefore judges and returns a different URL than the Worker for any spelling with a backslash after the ? or #.

Evidence (verified 2026-10-22, current main)

validate_crawl_url (Python) vs new URL(...).toString() (the Worker's serializer):

Found by the 470-case ASCII differential run for #618; this was the only remaining returned-URL mismatch class outside structural-delimiter spellings.

Why it matters

• The query is sent to the origin: ?q=a\b and ?q=a/b are different queries and can name different resources (search terms, signed parameters, Windows-style path parameters). The container crawls a URL the Worker would not, for a target both validators accepted — the same judged-vs-fetched divergence class as lab-ops-hub: container validator parses backslash authority differently from WHATWG (127.0.0.1\@example.com judged as example.com; example.com\@127.0.0.1 refused) #577 itself, in the opposite direction: lab-ops-hub: container validator parses backslash authority differently from WHATWG (127.0.0.1\@example.com judged as example.com; example.com\@127.0.0.1 refused) #577's fix overreaches past the components where the backslash is structural.
• The fragment arm changes the returned/recorded URL identity even though fragments are not fetched.

Acceptance criteria

• The \ → / replacement applies only up to the first ? or # (the scheme/authority/path portion); backslashes in the query and fragment pass through verbatim, matching the Worker byte-for-byte.
• The lab-ops-hub: container validator parses backslash authority differently from WHATWG (127.0.0.1\@example.com judged as example.com; example.com\@127.0.0.1 refused) #577 authority spellings still normalize and are still judged on the normalized host (http://127.0.0.1\@example.com/ refused as loopback, http://example.com\@127.0.0.1/ judged as example.com), and the existing backslash tests pass unchanged.
• New tests pin the query and fragment spellings in the table above.
• Python suite (test_container_validate.py) and TS suite (lib.test.ts) both green.

Suggested fix

Split the input at the first ?/# before the replacement and rejoin afterwards — or scope the existing raw.replace("\\", "/") to the head segment — taking care that a \ inside the authority is still replaced even when a ? appears later in the same spelling (it does: the head segment includes the full authority and path).