Surface config save failures in the GUI

`state.rs`'s `persist_and_reload` and ~9 sibling sites treat `save_atomic()` failure as `warn!` + continue: the UI keeps the in-memory value, the user sees success, and the setting silently reverts on restart (read-only config dir, disk full, …).

Policy proposal: a one-shot, non-blocking notification (gpui-component has a toast/notification primitive) the first time a save fails, per outage — not per keystroke. Keeping the in-memory value and skipping the agent reload on failure stays as is (that ordering is deliberate and documented).

Needs a new user-facing string, i.e. an ordered-position insert across all 20 locale files plus Crowdin seeding — which is why it was split out of #381.

Start in state.rs at persist_and_reload, then inspect the approximately nine sibling save_atomic call sites and the existing gpui-component toast or notification primitive. Add the user-facing string across all 20 locale files with Crowdin seeding, and verify that a failed save keeps the in-memory value, skips agent reload, and shows one non-blocking notification per outage rather than per keystroke.

Written by the indexing model from the issue text.