fix(navidrome-admin): resilient admin API calls + UI polish (#260)

* fix(navidrome-admin): force HTTP/1.1 + User-Agent + no idle pool for /auth and /api calls

Replaces plain reqwest::Client::new() in all nd_* Tauri commands
(navidrome_login, nd_list_users, nd_create_user, nd_update_user,
nd_delete_user, nd_list_libraries, nd_set_user_libraries) with a shared
nd_http_client() helper that:

- sets a real User-Agent (Psysonic/<version> (Tauri))
- pins HTTP/1.1 (avoids HTTP/2 ALPN that some reverse proxies abort on)
- disables the idle-connection pool so a second call doesn't reuse a
  TCP connection that the server or proxy has already half-closed
  (was producing intermittent "tls handshake eof" on external servers).

Adds nd_err() to flatten the reqwest error source chain into the
returned String, so the frontend surfaces the real cause (connection
refused, tls handshake eof, etc.) instead of reqwest's opaque
"error sending request for url (…)" wrapper.

* fix(navidrome-admin): retry + graceful error + server row polish

Rust (src-tauri/src/lib.rs):
- nd_http_client: HTTP/1.1, TLS 1.2 only, no idle pool — browser-parity
  for /auth and /api so strict reverse proxies that abort reqwest's
  default HTTP/2+TLS-1.3 handshake mid-flight get through.
- nd_retry: one retry after 500ms on connect/timeout errors only
  (ECONNRESET, TLS handshake EOF). Aggressive retries could push the
  nginx upstream probe into offline state; this is the minimal useful
  amount.
- nd_err: flatten the reqwest error source chain so the UI surfaces the
  real cause instead of reqwest's opaque wrapper.

User management UI:
- Sequential load (users, then libraries) instead of parallel, to avoid
  racing two TLS connections on a single nginx upstream slot.
- Friendlier failure state with a one-click Retry button that re-runs
  load(). Concrete Rust error is kept as muted sub-line.
- User row layout: Magic-String button sits consistently next to
  last-seen + delete for every user, regardless of admin status or
  library-name length. No more mid-row jitter.

Servers tab:
- "Use" button no longer redirects to Home; stays on the Servers tab so
  the active-badge migration is visible.
- Drag-and-drop reorder via grip handle (psyDnD), backed by a new
  setServers() action on authStore.
- Active server row now has an accent-tinted background on top of the
  border — harder to miss.

New i18n keys (userMgmtLoadFriendly, userMgmtRetry) added to all 8 locales.

---------

Co-authored-by: Psychotoxical <dev@psysonic.app>

src/locales/de.ts

@@ -481,6 +481,8 @@ export const deTranslation = {
     userMgmtDesc: 'Benutzer auf diesem Server verwalten. Erfordert Admin-Rechte.',
     userMgmtNoAdmin: 'Du brauchst Admin-Rechte, um Benutzer auf diesem Server zu verwalten.',
     userMgmtLoadError: 'Benutzer konnten nicht geladen werden.',
+    userMgmtLoadFriendly: 'Server hat nicht geantwortet — meistens ein Einzelfall.',
+    userMgmtRetry: 'Erneut versuchen',
     userMgmtEmpty: 'Keine Benutzer gefunden.',
     userMgmtYouBadge: 'Du',
     userMgmtAdminBadge: 'Admin',