[Bug]: SMB listings may omit long filenames and misparse directory rows #182

Problem

DBackup's SMB adapter delegates directory enumeration to samba-client (client.list()). With long backup archive filenames, smbclient can emit a line without the usual padded filename column. The downstream has observed cases where samba-client returns an empty or incomplete listing while a raw dir command still contains those entries.

Example (synthetic data):

  Critical_Backups_2026-10-10_01-20-00_full-000.tar A 2580549120  Sat Oct 10 01:20:00 2026
  nightly.index A 100  Sat Oct 10 01:20:00 2026
  snapshot-2026-10-10                Dn        0  Sat Oct 10 01:20:00 2026

Related downstream discoveries: inconsistent raw output type (string/Buffer), non-padded filenames, mixed-case DOS directory flags, and directories containing spaces. These are related symptoms of one logical SMB listing correctness problem, not separate feature requests.

Current upstream

At v4.0.0 / current main, src/lib/adapters/storage/smb.ts uses client.list(currentDir + "/*") for recursive listing and does not recover rows omitted by the library parser.

Proposed upstream-compatible fix

1. Normalize enumeration to a properly scoped current directory so names containing spaces are not interpreted as a glob/path outside the intended listing.
2. Recover otherwise omitted entries from the same smbclient directory output using a conservative right-anchored parser, retaining only syntactically valid rows.
3. Merge parsed entries by exact filename with ordinary list() output to avoid duplicates, preserving original API shape.
4. Recognize mixed-case DOS attribute flags and correctly distinguish files from directories.
5. Preserve existing path-prefix and credential-sanitization boundaries; no Homelab-specific paths or policy.
6. Include source-level regressions for long filename, partial/empty library listing, directory containing spaces, Dn attributes, and failed raw enumeration.

A maintained downstream has regression coverage for these cases. This issue is intended to converge that single logical patch with upstream rather than submit iterative repair commits as multiple unrelated PRs. A separate issue covers incomplete recursive listings after permission errors (#181), which has different compatibility considerations.