Cover more of S3 listing, and make its cases steer clients better

[tiptenbrink]

Makes the S3 listing cases say enough that a client can reach a correct and good implementation from the grading alone. Every S3 path is verified live against S3 and S3 Express.

New cases

• list-entry-properties: an entry's last_modified (RFC 3339, UTC, whole seconds) and storage_class, with one object in STANDARD_IA.
• list-fetch-owner: owner_id with fetch-owner=true. S3 honours only the literal true; the AWS C++ SDK sends 1 and fails.
• list-start-after and list-start-after-delimited. On a directory bucket, start-after is answered 501 NotSupported.
• list-all-pages, a walk over three pages, and list-over-1000, 1,001 keys at the default page size.
• list-truncated-without-token: a truncated page without a token must be reported as an error. It is offline only.
• list-prefix-special-characters: a prefix and a common prefix holding a space, +, &, =, % and é.
• list-bucket-missing on Express: CreateSession's 404 NoSuchBucket, and nothing more.
• list-other-delimiter now groups keys at -. Before, no key held a -, so a client that dropped the delimiter passed.

Grading changes

• A request has raw_query, the query as sent, and a check may carry because, which a failure reports.
• Every S3 profile fails a raw + in the query. S3 reads it as a space, so a continuation token or upload ID holding + breaks. The failure says to send %2B, or %20 for a space.
• Every S3 XML fixture starts with the XML declaration and escapes ETag quotes as ", as S3 sends them.
• The core s3/list-* cases accept encoding-type=url, as the operation cases do.
• Each S3 listing purpose states its rule. list-encoded-names says to ask for encoding-type=url: without it S3 writes U+0001 as , which XML 1.0 forbids, and a CR in a key reads back as LF.

Checks

cargo test passes, including a new test for the raw + check and its reason.