Remap wheel tilt and middle click, with dedicated GUI slots

[distractedhero]

Problem

Pointer input can't be remapped. The event tap subscribes to mouse and scroll
events, but eventCallback returns early for anything that isn't keyDown,
keyUp or flagsChanged, so a learned wheel binding is recorded and then
never acted on. Binding wheel tilt or middle click has no effect today.

Approach

HID identifies which physical device produced the input; CGEvent is where the
native action can be suppressed. The two are matched by hardware timestamp
within a 50ms tolerance. A matched pointer event is deferred 25ms rather than
blocked, since HID and CGEvent callbacks share the main run loop, and anything
that fails to match is replayed so unrelated input is never swallowed.

Direction comes from the signed HID pan value rather than the CGEvent delta, so
Natural Scrolling doesn't invert it.

A single physical tilt emits a burst of pan reports, so dispatch is debounced
per direction with a 150ms gesture gap.

Surfacing it in the UI

Wheel controls appear in the mapping window as slots 15, 16 and 17, named Wheel
Tilt Left, Wheel Tilt Right and Wheel Click, in their own card row alongside the
existing DPI slots. They ship unbound, since tilt and click report different HID
usages per device, so each user pairs them through the existing Learn Hardware
Trigger flow. Without these cards the capability would only be reachable by
hand-editing profiles.json.

They are independent slots, so binding the wheel no longer consumes a thumb
button mapping.

Verification

Tested on a Naga V2 HS over Bluetooth. Wheel tilt left and right and wheel click
all perform their mapped actions, native behaviour suppressed, no leakage. 136
HID pan reports collapsed to 14 dispatches across 14 physical tilts. Existing
tests pass, plus new coverage for the router, modifier release, and profile
decoding.

Logging

Per-event tracing is gated behind a NAGA_DEBUG environment variable. A normal
launch previously logged every HID report and mapping lookup; that now stays
silent unless the variable is set. Permission state, event tap state, profile
load counts and errors still log unconditionally.

🤖 Generated with Claude Code

https://claude.ai/code/session_017mrRRGrRu2XN8q5ruiZKkT

[DParent10]

This is excellent work, @distractedhero — thank you. Matching HID to CGEvent by hardware timestamp and deferring instead of blocking is exactly the right shape, taking direction from the signed HID pan so Natural Scrolling can't flip it is a detail I'd have missed, and the router tests are a real gift. I also want to call out a bonus you may not have noticed: because the router works for any slot, a learned mouse-button trigger now fires on buttons 1–12 too, which closes a limitation I'd documented in the user guide last week.

I pushed one small commit on top of yours (maintainer edit) rather than round-tripping:

• Menu bar title. Your AppDelegate hunk reverted the active-profile name in the status item, which landed in Fix profile switching from mouse buttons #12 a few days before your branch. I've restored it. If the title was too wide on your setup, say so and we can shorten it deliberately.
• Bindings lookup placement. deferMappedPointer runs for every tapped event including keystrokes, so I moved the hardwareBindingsForCurrentProfile() fetch behind the event-type checks.
• README troubleshooting now explains NAGA_DEBUG (open --env NAGA_DEBUG=1 …) and lists the lines the app actually prints today, since the old ones no longer exist.

One thing I'd like you to add to the PR description rather than change: the ButtonMapper changes — private event source for synthesized keys, key-up carrying the physical flags, and finishShortcut releasing modifiers that aren't physically held — touch every mapping the app sends, and the description doesn't mention them. I can see the motive (a mapped Cmd+Tab otherwise leaves the switcher open), and the tests cover it, but future readers will want that reasoning in the PR. A paragraph is plenty.

I'll do a hardware pass on my V2 HS today (tilt, click, Cmd+Tab, and a plain Cmd+C hold to check the modifier change) and merge once that's clean. Thanks again — this one's been on the wish list since #4.