Swipe into empty workspaces on Macs

[maralcbr]

Owner check on #629: the three-finger swipe worked with a window open but did nothing between empty workspaces. Hyprland 0.56 has no gesture option for this. Its default steps through existing workspaces only (m±1), and UnifiedWorkspaceSwipeGesture.cpp:74-75 refuses to swipe out of an empty workspace into a new one. gestures:workspace_swipe_use_r steps by number (r±1), so the swipe reaches empty and new workspaces, as Spaces do in macOS.

• apple-gestures.lua sets workspace_swipe_use_r = true when it registers Omarchy's Mac gesture, and only then. A user's own gesture, or omarchy_workspace_gesture = false, keeps Hyprland's stepping. It lives in the module so it moves with it into omarchy-mac's package (Hand the Apple desktop leaves to omarchy-mac's setup #631), and the runtime keeps no Mac code. The trade-off is that a user's use_r = false next to Omarchy's gesture is overridden; their own gesture is the way out.
• The Mac manual says the swipe reaches empty workspaces.
• x86 and Snapdragon are unchanged.

Tests: hyprland-apple-workspace-gesture-test.sh checks that Macs get use_r, that a user's own gesture or the opt-out doesn't, and that other platforms don't. It fails without the change. The related hyprland/touchpad suites pass.

Hardware, with a synthetic uinput 3-finger touchpad and the deployed config (no eval), starting from an empty workspace: the M2 went 5→6→7 (new, empty) and back 7→6→5→4 (with use_r set before or after the user files), and the M1 went 5→6→7→6→5→4 with this exact module. Swipes do nothing while the screen is locked (WorkspaceSwipeGesture.cpp:15), which explains some earlier "no movement" runs. hyprctl configerrors is empty on both.

Design debate: the numbering is sound (it skips other monitors' and bound workspaces, excludes specials, and r-1 from workspace 1 stays put). It suggested setting the value before the user files. That was adopted first, then moved into the module so the runtime stays free of Mac code under #631.

Second review (of the earlier placement): no blocking findings. It confirmed the r±1 semantics, that end() creates the missing workspace, and no regression on x86.

Update after #631 merged: I merged quattro-upstream into this branch. The four lines now sit in packages/omarchy-mac/share/omarchy/default/hypr/platform/apple-gestures.lua, and git carried them through the rename with no conflict. The #631 layout plus this change is deployed on both Macs: use_r true, gesture active, no config errors. On the unlocked M1, a synthetic swipe starting from an empty workspace went 5→6→7→6→5→3; workspace 4 was skipped, which fits r±1 jumping over a workspace held by its other monitor.