The end-of-session dialog no longer carries a second "finish" button.
"TERMINER TOUT" is now the only way out, and what it does with the target
depends on the AI training setting:
- setting on: the target is always sent, with a discreet reminder saying so;
- setting off: a "CONTRIBUER À L'IA" switch (off by default, never
remembered) decides whether the target is sent or kept on the device;
- banned account or failed analysis: neither switch nor upload.
Turning the switch on for the first time shows the full program disclaimer
(what is sent, pseudonymity, banning rules); refusing leaves it off. That
text moved to a shared showAiConsentDialog() so the settings switch and the
session switch cannot drift apart, and a new is_ai_terms_accepted flag
remembers the acceptance without silently enabling the global setting.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
