feat: gestures and software scroll inversion for MX Vertical - #783

Summary

MX Vertical can use neither gestures nor scroll inversion today, for two unrelated reasons.

Gestures. The device ships no dedicated gesture button, and the capture layer knows only two gesture sources (0x00c3, plus the MX Master 4 haptic panel). diag controls on real hardware:

device: MX Vertical Advanced Ergonomic Mouse (direct 046d:b020)
     cid    task   flags  capabilities
  0x0050  0x0038  0x0411  analytics-events
  0x0051  0x0039  0x0411  analytics-events
  0x0052  0x003a  0x0571  divertable, raw-xy, analytics-events
  0x0053  0x003c  0x0571  divertable, raw-xy, analytics-events
  0x0056  0x003e  0x0571  divertable, raw-xy, analytics-events
  0x00fd  0x00d2  0x0571  divertable, raw-xy, analytics-events
  0x00d7  0x00b4  0x03a0  divertable, raw-xy, force-raw-xy

There is no 0x00c3. 0x00d7 (virtual gesture button) is listed but never emits — armed with raw-XY it produced no events across repeated presses. 0x00fd, the DPI switch, is the control that actually fires, and it is raw-XY capable. Logitech's own asset metadata agrees: for this model it ships SLOT_NAME_GESTURE_{UP,DOWN,LEFT,RIGHT,CLICK}_BUTTON next to SLOT_NAME_DPI_BUTTON, i.e. it also treats the DPI button as the control that gestures here.

The GUI had a second, independent problem: map_slot_name knew only SLOT_NAME_MODESHIFT_BUTTON, so this device's DPI button produced no hotspot and could not be bound at all.

Scroll inversion. Capabilities::scroll_inversion is derived from 0x2121 (HiResWheel). MX Vertical reports 30 features and none is 0x2121, so the toggle read "Unavailable" permanently with no way to reach the setting.

Changes

openlogi-device — the DPI/ModeShift family joins GESTURE_SOURCE_BUTTONS. Also guards the capture session's DPI pass against re-arming a CID already armed with raw-XY: that write carries none and would strip the reporting the hold depends on, reachable exactly on a device whose DPI button is its gesture source.

openlogi-core — ButtonId::DpiToggle is now a HID++ gesture source. DPI cycling stays the out-of-the-box behavior: default_binding_for returns Single for that button, and a Single shape drops out of the gesture-map lookup.

openlogi-hook — new EventDisposition::InvertScroll. macOS negates the CGEvent delta fields in place, so the event keeps its phase, momentum and pixel precision and no synthetic replacement re-enters the tap. Only fields the event actually carries are negated, since writing an axis the device left empty would introduce a delta an app reading that field would honour. evdev and WH_MOUSE_LL have no rewrite path yet and pass the disposition through rather than swallowing the scroll.

openlogi-agent-core — HookMaps::invert_scroll carries the identities the hook rewrites, built from the config while excluding natively-capable devices (their setting goes to the firmware; rewriting on top would invert twice). Trackpad scroll is never touched, since macOS already applies its own natural-scrolling preference. Inversion is keyed per device rather than scoped to the selected one, so disabling the selected device empties the button maps without stopping another device's inversion. Two identical directly-attached mice share the one vendor/product pair the hook can see, so when their settings disagree the identity is dropped with a warning rather than applied to both.

openlogi-desktop — SLOT_NAME_DPI_BUTTON maps to ButtonId::DpiToggle (the five per-direction gesture markers stay unmapped: same control, and the hotspot builder does not dedupe). The scroll-inversion toggle is offered for pointer-capable devices on macOS, still gated on the native capability elsewhere.

Testing

Commands run on the final tree, all clean:

cargo fmt --all -- --check
cargo clippy -p openlogi-core -p openlogi-device -p openlogi-hid -p openlogi-hook -p openlogi-agent-core -p openlogi-desktop --all-targets -- -D warnings
cargo test -p openlogi-core -p openlogi-device -p openlogi-hid -p openlogi-hook -p openlogi-agent-core -p openlogi-desktop
RUSTDOCFLAGS="-D warnings" cargo doc --workspace --no-deps --document-private-items --exclude openlogi-ui --exclude openlogi-desktop --exclude openlogi-overlay --exclude openlogi-agent

646 tests pass. Eight new ones cover: DPI CIDs resolving to ButtonId::DpiToggle, the Single default that keeps DPI cycling, both DPI slot names, the deliberately unmapped direction markers, exclusion of natively-capable and receiver-paired devices, the per-device inversion property, and the identical-device disagreement case.

Hardware verification — runtime-tested on MX Vertical (Bluetooth-direct, macOS):

Not runtime-tested on Linux or Windows — those hook arms are compile-only changes that pass the new disposition through. Not tested with two identical mice attached; that path is covered by unit tests only.

Rebased onto master after the openlogi-device extraction and the hook backend-trait refactor; the only conflict was in macos.rs, where negate_scroll_axis now sits beside usable_scroll_delta and start() stays where the trait refactor put it.