feat: add mouse buttons 6–9 as pickable actions

[MichaelDanCurtis]

Summary

Adds four pickable actions — MouseButton6–MouseButton9 — that synthesize
mouse buttons 6–9, for apps/games/CAD that bind buttons beyond the standard
five. Mirrors the existing MouseBack/MouseForward pattern.

Why: the Action vocabulary capped output at "button 5". Users wanting a
physical button to emit button 6+ (Blender, MMO-mouse emulators, etc.) had no
path, even though the underlying injection layer already supported arbitrary
button numbers on macOS/Linux.

What changes

Layer	Change
openlogi-core/src/binding.rs	+4 Action variants; wired into label() ("Button 6"…), category() (Category::Mouse), and catalog()
openlogi-inject/src/inject.rs	macOS → post_other_button(5..=8); Linux → evdev BTN_FORWARD/BACK/TASK/0; Windows → log-and-skip
openlogi-gui/src/mouse_model/picker.rs	icon arms (compiler-forced — exhaustive match, no wildcard)
openlogi-gui/locales/*.yml (×20)	"Button 6"–"Button 9" translation keys, each locale's own "button" word

Purely additive — no schema_version bump, no migration. New unit variants
serialize as bare TOML strings ("MouseButton6"), consistent with existing
extra-button actions.

Platform coverage

• macOS — full (existing post_other_button already takes any number)
• Linux — full (evdev BTN_* family)
• Windows — log-and-skip. SendInput's mouse path carries flags for
  buttons 1–5 only; there's no flag for 6+. Documented gap, matches the
  existing "no platform equivalent → debug log + skip" pattern. Left out of
  scope to fix here since Windows is a preview target.

Test plan — what was verified locally

• cargo test -p openlogi-core — 84 pass, including new
  extra_mouse_button_labels and extended category_mouse_variants;
  all_catalog_variants_roundtrip_toml now exercises the 4 new variants
• cargo test -p openlogi-inject — 1 pass
• cargo test -p openlogi-agent-core — 33 pass (consumes the Action
  enum; confirms no downstream missing-match arms)
• cargo build -p openlogi-gui (macOS) — succeeds after installing the
  Metal Toolchain; the picker's exhaustive match compiles, proving all
  four variants are covered
• Ran the dev-built GUI on real hardware against an MX Master 3S over
  Bluetooth-direct — confirmed Button 6/7/8/9 appear in the MOUSE section
  of the action picker, with icons, after Forward (Button 5)
• cargo build -p openlogi-inject (macOS) — succeeds
• Linux/Windows cargo check --target — no cross-targets installed
  locally; relies on CI. evdev 0.13.2 confirmed to expose
  BTN_FORWARD/BTN_BACK/BTN_TASK/BTN_0.

Honest note on emission coverage

Button reception by a downstream app (e.g. Blender registering MB6) was not
end-to-end tested. The emission path reuses the already-shipping
post_other_button (proven for buttons 3/4) with a different integer argument
(5–8) on the same MOUSE_EVENT_BUTTON_NUMBER field, so the risk is low — but
flagging it rather than claiming full end-to-end verification.

Notes

• Naming: variants MouseButton6–MouseButton9, labels "Button 6"–
  "Button 9". Unlike Back/Forward these have no universal semantic meaning.
• Spec & plan live in docs/superpowers/specs/2026-06-29-mouse-buttons-6-9-design.md
  and docs/superpowers/plans/2026-06-29-mouse-buttons-6-9.md (in the last two
  commits on this branch). Happy to rebase them out if you'd prefer code-only.

[greptile-apps]

Greptile Summary

This PR adds pickable actions for mouse buttons 6 through 9. The main changes are:

• New Action variants, labels, categories, catalog entries, and core tests.
• GUI picker icon handling and locale keys for the new actions.
• macOS and Linux injection paths for the new buttons.
• Windows debug-and-skip handling for unsupported buttons.

Confidence Score: 4/5

This is close, but the Linux Button 9 mapping should be fixed before merging.

• Linux now advertises the new emitted button codes.
• MouseButton9 still emits BTN_0, so apps can receive the wrong numbered button event.
• The core, GUI, macOS, and Windows changes look consistent with the intended additive behavior.

Files Needing Attention: crates/openlogi-inject/src/inject/linux.rs

Important Files Changed

Filename	Overview
crates/openlogi-core/src/binding.rs	Adds the new action variants with labels, mouse categorization, catalog entries, and focused tests.
crates/openlogi-inject/src/inject/linux.rs	Adds Linux emission and uinput capabilities for the new actions, but Button 9 still maps to the wrong evdev code.
crates/openlogi-inject/src/inject/macos.rs	Maps the new actions to the existing macOS other-button event path.
crates/openlogi-inject/src/inject/windows.rs	Adds explicit debug-and-skip handling for unsupported Windows output.
crates/openlogi-gui/src/mouse_model/picker.rs	Adds picker icon coverage for the new mouse-button actions.

_{Reviews (3): Last reviewed commit: "feat(gui): add Button 6-9 translation ke..." | Re-trigger Greptile}

[greptile-apps]

Linux Button Numbers Are Reordered

The new Linux mapping mixes semantic side-button codes with the positional BTN_0 range, so the emitted events do not line up with the user-facing button numbers. A Linux app that binds the numbered extra buttons can receive BTN_FORWARD, BTN_BACK, BTN_TASK, and then BTN_0 for Button 9, which makes Button 9 look like the first generic button instead of the ninth numbered button.

[MichaelDanCurtis]

Closing for now — withdrawing until the feature is fully built and tested locally. Sorry for the noise.

[greptile-apps]

Button 9 misfires

MouseButton9 still emits BTN_0 on Linux. That code is the first entry in the generic BTN_0..BTN_9 range, not a ninth positional mouse button, so an app that binds numbered extra buttons can receive this as the first generic button instead of the selected Button 9 action. The new capability list now advertises this event, which means the wrong event is delivered instead of being dropped.