Fix AI assistant crash when the AI server errors mid-stream

[elias-ba]

Description

The AI assistant talks to the Apollo AI server over a streaming connection. When that server replied with an error status, the code that handles errors assumed it was dealing with an ordinary JSON response and tried to read a field off it. Over a streaming connection the response body isn't a plain object, so this crashed, the background job was discarded, and the real error was hidden from the user.

This fixes the error handler so an error response over a streaming connection now returns a clean error message instead of crashing. The failing HTTP status is also included in the log so the underlying cause is easier to trace.

Closes #4941

Validation steps

1. With the AI assistant enabled, open the assistant on a job (or the workflow chat) in a project you can edit.
2. Put the AI server into a failing state so a request comes back unsuccessful, for example point the Apollo endpoint at an unreachable or misconfigured URL.
3. Send a message to the assistant.
4. Confirm the assistant shows a clean error message and the chat stays usable: the message should not hang on a loading state and the editor session should not drop. Before this fix, an error returned mid-stream crashed the handler and the message silently failed.

Additional notes for the reviewer

AI Usage

Please disclose whether you've used AI anywhere in this PR (it's cool, we just
want to know!):

• I have used Claude Code
• I have used another model
• I have not used AI

You can read more details in our
Responsible AI Policy

Pre-submission checklist

• I have performed an AI review of my code (we recommend using /review with Claude Code)
• I have implemented and tested all related authorization policies. (e.g., :owner, :admin, :editor, :viewer)
• I have updated the changelog.
• I have ticked a box in "AI usage" in this PR

[github-actions]

Security Review ✅

• S0 (project scoping): N/A — the change only alters how an error message is extracted from an HTTP response body in lib/lightning/ai_assistant/ai_assistant.ex:1437; no queries or data-access paths were introduced or modified.
• S1 (authorization): N/A — no new controller actions, LiveView events, or channel handlers; the diff refactors a private helper (error_message_from_body/1) inside an already-gated flow.
• S2 (audit trail): N/A — no config-resource writes (workflows, credentials, project settings, etc.); the change touches only error-response handling.