fix: guard against None message before accessing .content

[qizwiz]

What

Extend two guard conditions in app/llm.py to also check choices[0].message is None before accessing .content.

Why

The existing guards:

if not response.choices or not response.choices[0].message.content:

protect against an empty choices list and empty/None content, but if choices[0].message is None (returned by some providers on content-filtered responses with HTTP 200, e.g. Gemini PROHIBITED_CONTENT), accessing .content raises AttributeError: 'NoneType' object has no attribute 'content' before the guard evaluates.

Fix

# Before:
if not response.choices or not response.choices[0].message.content:

# After:
if not response.choices or response.choices[0].message is None or not response.choices[0].message.content:

Both occurrences (~line 425 and ~line 595) are fixed. No behaviour change for well-formed responses.