Quiet honesty

Clearer reasons when local AI runs in its lighter mode, and an honest note on the faster-model experiment we chose not to ship.

• Clearer, honest messaging when local AI runs in its lighter mode. If the app cannot load its local AI model, it now names the real reason instead of guessing. When no model is installed it says so and points you to Settings to add one, and it only mentions low memory when a model really is present but will not fit right now
• We tested a smaller, faster local AI model for low-memory PCs and chose not to ship it. In blind side-by-side testing it gave worse and sometimes unsafe advice, and for a repair tool a fast wrong answer is worse than a slower right one, so we kept the higher-quality model. Faster local AI for those PCs is still coming, through graphics-card acceleration