fix(ssd): keep frames on-screen below panels; refresh cursor leaving borders

[ebenali]

Follow-up to the SSD work, fixing two issues you hit with Waybar (which runs in the bottom layer with a 30px top exclusive zone).

1. SSD frame placed off-screen / under the panel

Placement positions the client inside the usable area, but at map time the frame doesn't exist yet (insets are zero). So a client at the area's top-left left its border off the left edge (x<0) and its titlebar up inside the panel's reserved zone — exactly the "client at x=0, left border not shown, y a few pixels from the top" you described. constrain_to_usable only reserves panel-inset edges, not the output bounds, so the left border stayed clipped.

Fix: build the frame before placement/logging and add clamp_frame_into_area() — once the insets are known, pull the whole frame (client + insets) inside the placement area so every border is on-screen and the titlebar sits flush below the panel. The mapped geometry is logged after the clamp so restore geometry matches.

With real Waybar, foot now maps at +1+54: client x=1 → left border at x=0 on-screen; frame top = 54−24 = 30, exactly the panel's bottom edge. No overlap, nothing clipped.

2. Resize cursor stuck when moving back into the window

Hovering a border/corner set a resize xcursor and returned without releasing the client's pointer focus, so moving back inside the same surface sent no pointer-enter and the client never reasserted its cursor — it kept the resize shape. Fix: clear the client's pointer focus while the cursor is over our decoration; re-entering the client re-sends enter and it restores its own cursor.

Testing

• Full suite 27 green (g++/ASan/UBSan), incl. integration_window_state (restore geometry now matches the clamped mapped position) and integration_ssd; clang + release clean.
• Headless with real Waybar confirmed the on-screen, below-panel placement.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com