feat(app): settings grouped by purpose, and four defects an audit found

[Helldez]

Settings

Each category now shows the recommended configuration first and folds the rest into a collapsed Experimental group. That is a statement about evidence rather than about how finished the code is: inside are the levers measured on one device, measured once, or still owed a measurement. They stay in the release build on purpose, because testing them on hardware other than the one test phone is what this app is for, and a lever nobody can reach is a lever nobody can refute. The caveat is stated once in the group header instead of leaking into some descriptions and not others.

Every description was rewritten to say what the setting does for the person reading it:

Before	After
\mmap baseline (no streaming)\	Load the whole model instead
\Predictive prefetch (experimental)\ … right ~85% of the time	Ask the next layer what it wants
\Pinned (dma-buf)\	Pinned
\Active experts (top-k)\	Experts per token

Out went the measured figures, which need the device, the model and the day beside them to mean anything and have none of that room under a switch. Out went the implementation names with them: \O_DIRECT, top-k, dma-buf, mmap and KV cache are not what someone deciding whether to turn something on needs to know. The metrics screen keeps the flag names deliberately, because there the reader is matching the UI against a CSV column and the technical name is the vocabulary.

Four defects, all found by auditing rather than by something failing

The session signature is derived from the argv instead of being a hand-written list beside it. The two had to be kept in step with nothing enforcing it, and omitting a field is a silent bug: the setting appears to change while the engine keeps running the old configuration. Three of the four merges in this release collided on exactly that list.

A malformed end-of-turn summary no longer strands the UI. The whole handler sat inside a catch with no failure branch, so a parse error left the state in GENERATING with no turn committed and nothing said. The streamed answer is now kept, the reason is shown, and the state returns to READY.

\MainActivity\ drops from about 1050 lines to under 700. The model download and import UI moves to \ModelPickerUi.kt, which shares nothing with the chat screen. No logic moved, only its address.

Dead code removed: a field whose own comment described a use it did not have, two functions nobody called, and five string resources describing a UI two rewrites ago.