fix(code): continue listings after skipped candidate-only pages #131

When list_files reaches its candidate scan cap but every candidate is skipped during canonical/symlink/file verification, it can return truncated: true with no paths and no nextAfterPath. The negotiated pagination validator rejects that result, so valid files after a leading run of skipped candidates cannot be reached.

Please make the scan continue until it produces a valid page/cursor, exhausts candidates, or hits the existing deadline, and add a regression test with more than one candidate window of skipped entries followed by a valid file.

Found while mirroring #106 into ClickHouse/ai.