[Bug] Handle Gemini error responses safely in /quick ai

[CodeMatrix1]

What happened?

Description

When /quick ai is used with Gemini and the provider returns an error payload instead of generated candidate text, the AI flow fails.

The current implementation assumes the Gemini response always contains text at:

response.data.candidates[0].content.parts[0].text

In practice, Gemini can return an error object instead of candidates, for example when the model is overloaded or temporarily unavailable. In those cases, /quick ai does not handle the response shape safely.

Expected behavior

• /quick ai should safely handle Gemini responses that do not include candidates
• if Gemini returns an explicit error payload, the app should surface the provider message instead of failing
• if Gemini returns an unexpected shape without generated text, the app should log the error in gemini response.

Actual behaviour

• /quick ai assumes the Gemini response always contains generated text at response.data.candidates[0].content.parts[0].text
• when Gemini returns an error payload instead of candidates, the AI flow fails
• proper logs are not generated
• the modal falls back to a generic failure message instead of handling the provider response safely

Steps to reproduce

1. Install the Quick Replies app
2. Configure Gemini as the AI provider
3. Trigger /quick ai
4. Enter a prompt and click Generate
5. Hit a Gemini error response such as overload / temporary unavailability
6. Observe that the response does not contain generated candidate text

Relevant ScreenShots

Version

v8.2.1

What browsers are you seeing the problem on?

Firefox

[CodeMatrix1]

@iam-vipin Sorry for the tag. I realized my earlier PR had drifted and overlapped with changes that were already merged, so I narrowed it down to just this specific Gemini /quick ai issue. It should be reviewable now.