Qwen starts faster and fullscreen dictation feels smoother

This update makes Mallo feel better in the first few minutes and the first few seconds. Qwen gets ready much faster on a cold start, onboarding points new users toward a better default model path, and the HUD behaves more naturally over fullscreen apps.

Qwen Starts Much Faster

One of the biggest quality improvements here is cold start time. Qwen now loads much faster, so the first dictation after launch feels less like waiting for the app to catch up.

That matters most for people using Mallo as a quick utility. If the first request is slow, the app feels heavy. This pass makes Qwen feel much closer to the instant tool it should be.

Better Onboarding Model Choice

Onboarding now makes a better default model choice instead of pushing new users through an older first-run path by default.

The goal is simple: less setup ambiguity, less second-guessing, and a stronger first impression for people who just want to install Mallo and start dictating.

A Better HUD Over Fullscreen Apps

The dictation HUD now behaves better around fullscreen apps, which removes some of the awkwardness that could show up when voice input had to coexist with a focused fullscreen workspace.

This is the kind of polish that helps Mallo disappear into the flow. When the HUD stays where it should and feels more reliable over fullscreen windows, dictation feels less disruptive.