fix(core,agent): stop force-resetting native scroll inversion on reconnect

[4ni1ak]

Summary

• DeviceConfig::invert_scroll is a bare bool (unlike dpi/smartshift,
  which are Option<T>), so a device-level false is indistinguishable
  from "never touched by the user".
• configured_wheel_mode() in the orchestrator didn't account for that
  ambiguity: it resolved the reapply target's inverted value to
  Some(effective_invert_scroll(..)) for any device that merely exposes
  the native scroll-inversion capability, regardless of whether the user
  had ever opened OpenLogi's scroll-direction setting.
• Because the agent's in-memory device list is empty on every process
  start — which happens on every host reboot — every online device is
  treated as newly-sighted at boot, so the reapply-on-reconnect path
  (reapply_volatile_settings / apply_native_wheel_modes) unconditionally
  wrote the native HID++ invert bit to false on any capable mouse on every
  reboot, silently overwriting whatever inversion state the device was
  actually in.

This matches what #1205 reports: settings the user never configured in
OpenLogi getting reset after every reboot.

The codebase already recognized this exact ambiguity in the legacy-config
migration fold (config/identity.rs, "invert_scroll is a bare bool, so
false is indistinguishable from never-set") but that care wasn't applied
to the reapply path.

Changes

• crates/openlogi-core: add DeviceConfig::configured_invert_scroll(),
  returning None unless there is real evidence of user intent — a link
  override (Option<bool>, so it can carry a deliberate false) or a
  device-level true. Unit tests cover the never-touched, device-level-true,
  and link-override-wins-even-when-false cases.
• crates/openlogi-agent-core: configured_wheel_mode() now calls
  configured_invert_scroll() instead of effective_invert_scroll(), so an
  untouched device's native inversion is left alone on reconnect/reboot —
  matching the "None means preserve the device's current value" contract
  the function's own doc comment already states, and that dpi/smartshift
  already honored. Added a regression test
  (configured_wheel_mode_leaves_unset_inversion_unmanaged) for the
  never-configured + capability-present case.

Testing

• cargo test -p openlogi-core configured_invert_scroll — new unit tests, pass.
• cargo test -p openlogi-agent-core configured_wheel_mode — existing +
  new regression test, pass.
• cargo fmt --all -- --check — clean.
• cargo clippy --all-targets -- -D warnings for the full affected package
  set (openlogi-core, openlogi-agent-core, openlogi-agent,
  openlogi-cli, openlogi-desktop, openlogi-device, openlogi-hid,
  openlogi-hook, openlogi-inject, openlogi-ipc, openlogi-overlay,
  openlogi-permissions, openlogi-ui, openlogi, xtask) — clean.
• cargo test for that same affected package set — all pass.
• Not runtime-tested on real hardware. This bug only manifests across an
  actual host reboot with a device whose native inversion state was set
  outside OpenLogi (or persisted from a prior host); that isn't something
  that can be verified from this environment. Reviewers with a mouse
  exposing HID++ 0x2121 invert bit (e.g. MX Master 3) can verify by: (1)
  setting the mouse's native scroll direction via another tool/host without
  ever touching OpenLogi's scroll-direction toggle, (2) confirming
  config.toml has no invert_scroll = true and no per-link
  invert_scroll override for that device, (3) rebooting with OpenLogi
  running, and (4) confirming the device's native inversion state is
  unchanged (pre-fix, it would reset to non-inverted).

Fixes #1205

[greptile-apps]

Greptile Summary

The PR stops reconnect handling from treating an ambiguous device-level invert_scroll = false as an instruction to overwrite native device state.

• Adds an intent-aware inversion accessor that preserves explicit per-link overrides.
• Uses that accessor when resolving reconnect and configuration-reload wheel settings.
• Adds regression coverage for untouched, enabled, and explicitly disabled inversion states.
• Serializes macOS workspace-notification tests that share the process-global notification center.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
crates/openlogi-core/src/config/device.rs	Adds intent-aware inversion resolution while retaining the existing effective-value accessor.
crates/openlogi-agent-core/src/orchestrator.rs	Uses configured inversion intent when determining which native wheel fields to reapply.
crates/openlogi-agent-core/src/orchestrator/tests.rs	Adds regression coverage ensuring untouched inversion remains unmanaged.
crates/openlogi-core/src/config/tests.rs	Covers untouched inversion, device-level true, and explicit per-link false precedence.
crates/openlogi-agent/src/tray.rs	Serializes the two tests that post process-global macOS workspace notifications.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Device reconnect or config reload] --> B[Resolve current route and capabilities]
    B --> C{Explicit link override?}
    C -->|Yes| D[Reapply explicit true or false]
    C -->|No| E{Device-level inversion true?}
    E -->|Yes| F[Reapply inversion enabled]
    E -->|No| G[Leave native inversion unchanged]
    D --> H[Write supported wheel fields]
    F --> H

Loading

_{Reviews (4): Last reviewed commit: "fix(agent): serialize the macOS tray tes..." | Re-trigger Greptile}