fix(cua-driver): isolate Wayland cursors across outputs

[jacob-vincent-mink]

Scope

• create one click-through layer-shell surface per enabled Wayland output
• route global cursor coordinates into deterministic output-local coordinates
• clear the previous surface when cursors move, hide, or end
• keep named session render state and lifecycle tombstones independent
• commit the first transparent buffer on every configured or hotplugged surface
• preserve demand-driven idle scheduling
• adapt the extracted work to current implicit-session revival semantics

This PR is intentionally limited to the native Wayland cursor overlay. It does not add Hyprland window discovery, capture, or input routing; browser cursor visualization; or the negative-coordinate placement-state change from #3019.

Existing work and coordination

• Fixes cua-driver: native Wayland cursor overlay loses output and session identity #3150.
• Refs e2e(cua-driver): require visible pointer and badge on Linux, Windows, and macOS #2879 for the broader cross-platform visible pointer and badge oracle.
• PR feat(cua-driver): add multi-display cursor support for macos #3019 touches wayland/overlay.rs for a distinct negative-coordinate fix. This draft should be rebased after whichever same-file change lands first; it does not supersede that PR.
• Hyprland capture and input remain owned by fix(cua-driver): fail closed on unproven Wayland window capture #2964 and stacked fix(cua-driver): support Hyprland capture and input #3052.

The first three commits were moved with git cherry-pick -x from spencerbull/cua, preserving Spencer Bull as their author and recording each source SHA. The final commit is the current-main lifecycle adaptation.

Validation

Candidate: dbb306c3c7fad34e3fdb3444a2309f1f4d14f46b

• git diff --check refs/remotes/upstream/main...HEAD
• cargo fmt --all -- --check
• cargo test -p platform-linux wayland::overlay --lib: 18 passed
• cargo test -p platform-linux --lib: 278 passed, 4 environment-dependent ignored
• cargo test -p cua-driver --test wayland_overlay_idle_test: compiled; 2 live-display tests ignored without an isolated Wayland runtime
• cargo check -p platform-linux --all-targets

Known gap before readiness

The extracted candidate has not yet been replayed through the canonical Linux desktop E2E harness on a current-main Hyprland or Sway runtime. The original source commits were live-tested on a three-output Hyprland session, including capture completion with every overlay layer active, but that earlier source SHA is supporting evidence rather than certification of this rebased candidate.

[jacob-vincent-mink]

Attribution coordination: the three extracted commits are authored by @spencerbull and retain cherry-pick -x source SHAs, but the repository attribution check requires a distinct same-repository source PR whenever the landing PR author differs from the preserved commit author. These commits came from a fork branch with no upstream source PR, so there is no truthful Salvaged from #... reference I can add. I will keep this draft blocked rather than rewrite away Spencer’s authorship. @spencerbull, could you advise whether you want to sponsor/open the focused contribution or establish an upstream source PR that preserves the provenance? The implementation and scope can remain unchanged.