Keep Google-only accounts out of the M365 scan

Graph returned 404 for /users/google:<email>/mailFolders: Google-
Workspace-only accounts (id "google:<email>") reached the M365 scan.

- getSelectedUsers() fell back to "every ticked checkbox in the DOM"
  whenever no M365 account was ticked — including Google-only rows. The
  fallback is now only used before the user list has loaded, and never
  includes google: rows.
- startScan() skips the M365 part (with a log line) when M365 sources are
  ticked but no M365 accounts are: the server treats an empty account
  list as "every user in the tenant".
- run_scan: _resolve_manual_users -> _resolve_scan_users, which now also
  drops google: ids (reported as a phase message). Covers scheduled scans
  and profiles that already contain Google-only accounts. Never widens to
  all users.

Tests: 3 new in tests/test_manual_users.py; the run_scan one fails on the
old engine with exactly the reported Graph call.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

static/js/CLAUDE.md

@@ -91,7 +91,7 @@ Never revert to `!!window._googleConnected` / `_fileSources.length > 0` — thos
   - The profile editor's "+ Tilføj konto" looks the email up via `/api/users/lookup` and stores `{id: <directory id>, displayName, email, manual: true}`. It falls back to `"manual:<email>"` only when the lookup fails. Checkboxes with `data-manual` save as objects; everything else saves as plain id strings.
   - `_applyProfile()` → `_addProfileManualUsers()` pushes manual accounts into `S._allUsers` (`manual: true`, `platform: 'm365'`). Legacy `manual:<email>` entries are looked up and the pending ID is rewritten to the resolved one. Without this the account can never be ticked in the sidebar, so it's never scanned.
   - `_userListFinal()` keys off `S._m365UsersLoaded`, not `S._allUsers.length`, because manual users can be pushed before the directory loads.
-  - Server side, `scan_engine._resolve_manual_users()` resolves any remaining `manual:<email>` (e.g. from scheduled scans) via `conn.lookup_user()`. It skips and reports failures and never widens to "all users".
+  - Server side, `scan_engine._resolve_scan_users()` resolves any remaining `manual:<email>` (e.g. from scheduled scans) via `conn.lookup_user()`, and drops `google:<email>` (Google-only accounts, which Graph 404s on). It skips and reports failures and never widens to "all users".
 - **Pending profile users wait for the Google merge** — `_applyPendingProfileUsers()` keeps `window._pendingProfileUserIds` until every ID is matched or `_userListFinal()` (users loaded, and Google either not connected or `S._googleUsersMerged`). `_mergeGoogleUsers()` re-applies in its `finally`. A saved `google:<email>` also matches an M365 user merged to `platform: 'both'` via `u.googleEmail`. Do not clear the pending IDs unconditionally — Google-only users would stay unticked.
 - **Date presets** — stored as `years * 365` (integer days). Do not use `* 365.25`.
 - **`copyTokenLink` is async** — called from `onclick` as fire-and-forget. Do not make it synchronous.