The error box said only "Error loading the AI summary". The status of
the failed request was raised and then dropped, so a user could not tell
an LLM server that is unreachable (502) from a plugin that is switched
off (403) or a request that was rejected (400).
The reason is now appended to the message. On a phone or in a webview,
where dev tools are not an option, it is the only thing the person
looking at the failure has to go on.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
