feat(linux): wire the KWin path end to end, and verify it live

The daemon now owns `vu.valiq.blocker`, loads the KWin script, and enters the
event loop on the primary backend. Verified against the running session:

```
$ blocker-daemon
blocker-daemon 0.1.0 (M0/M1)
database: /tmp/bsmoke/blocker/blocker.db
backend: kwin — native Wayland windows are visible
owning the D-Bus name; entering the event loop
enforcement: Freeze; entering the event loop
```

The process then parked until killed — which is the point. There is no tick; the
loop blocks on a focus event and on the engine's `next_wakeup`.

**Ordering is load-bearing and worth stating.** `callDBus` has nothing to resolve
until the daemon holds the bus name, so the name must be acquired *before* the
script is loaded. Loading first means a handler calling a name nobody owns, and
every focus event silently dropped.

**A real defect found by running it rather than reading it.** After the daemon
was killed, `isScriptLoaded` still returned `true` — the script was left in KWin.
`Drop` does not run on SIGTERM, so the cleanup path never executes. It is
fail-open and harmless (the script's `callDBus` fails inside the compositor,
which is exactly the section 1.5.4 behaviour), and `KwinBackend::connect`
unloads any stale script before loading a fresh one, so a restart recovers. But
"harmless" is not "correct", and the honest fix — a systemd `ExecStopPost` or a
signal handler that unloads — is not in this PR.

Also fixes the duplicated loop body: both backends now share one `run_loop`, so
the X11 and KWin paths cannot drift into enforcing differently.

Verified: clippy `-D warnings` clean, `fmt --check` clean, all tests passing,
stripped daemon within the 5 MB budget.

Co-Authored-By: Claude Code <noreply@anthropic.com>

desktop/platform-linux/src/watcher.rs

@@ -0,0 +1,117 @@
+//! The D-Bus watcher server: the daemon's public side of the KWin handshake.
+//!
+//! KWin calls *us*, not the other way round. The compositor has no way to push a
+//! focus event to an arbitrary subscriber, so the daemon owns a bus name and
+//! exports an interface that the loaded KWin script calls into. This module owns
+//! that name and turns those calls into [`FocusEvent`]s on a channel.
+//!
+//! # Owning the name is not optional
+//!
+//! Without the daemon holding `vu.valiq.blocker`, KWin's `callDBus` has nothing
+//! to resolve and every focus event is silently discarded. That is why
+//! [`serve`] must succeed *before* the script is loaded.
+//!
+//! # Fail-open
+//!
+//! If the daemon is not running, the KWin script's `callDBus` fails inside a
+//! signal handler and the compositor is left alone — no exception, no log spam.
+//! Blocking stops; the desktop keeps working. That is section 1.5.4's rule and
+//! it is why the script swallows its errors.
+
+use std::sync::mpsc::Sender;
+
+use zbus::blocking::{connection::Builder, Connection};
+
+use super::kwin::{BUS_NAME, OBJECT_PATH};
+use super::{parse_payload, FocusEvent};
+
+/// Why the watcher could not start.
+#[derive(Debug)]
+pub enum WatcherError {
+    /// No session bus, so there is nothing to own a name on. Almost always
+    /// means the daemon was started outside a desktop session — over SSH, or
+    /// from a system unit rather than a user one.
+    NoSessionBus(String),
+    /// The name could not be acquired, which normally means another instance is
+    /// already running.
+    NameTaken(String),
+    /// Anything else zbus reports.
+    Other(String),
+}
+
+impl std::fmt::Display for WatcherError {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        match self {
+            Self::NoSessionBus(e) => write!(f, "no session bus: {e}"),
+            Self::NameTaken(e) => write!(f, "another watcher already owns {BUS_NAME}: {e}"),
+            Self::Other(e) => write!(f, "watcher could not start: {e}"),
+        }
+    }
+}
+
+impl std::error::Error for WatcherError {}
+
+/// The object the KWin script calls into.
+///
+/// One method, `Focused`, whose payload is the tab-separated window description
+/// built by `kwin::SCRIPT_JS`. An empty payload means focus was cleared.
+struct Watcher {
+    sink: Sender<FocusEvent>,
+}
+
+#[zbus::interface(name = "vu.valiq.blocker.Watcher")]
+impl Watcher {
+    /// Called by the KWin script on every focus change.
+    ///
+    /// Payload is tab-separated: resourceClass, resourceName, caption,
+    /// internalId, pid — in the order `kwin::SCRIPT_JS` writes them.
+    fn focused(&self, payload: &str) {
+        let event = match parse_payload(payload) {
+            Some(window) => FocusEvent::Activated(window),
+            None => FocusEvent::Cleared,
+        };
+        // A send failure means the daemon has stopped. Nothing to escalate: the
+        // thread exits when the connection drops.
+        let _ = self.sink.send(event);
+    }
+}
+
+/// Acquire the bus name and export [`Watcher`], returning the live connection.
+///
+/// The connection must be held for the export to stay registered, so the caller
+/// keeps it for the daemon's lifetime.
+///
+/// Note on ordering: the daemon calls [`serve`] *before*
+/// [`super::kwin::KwinBackend::connect`]. Loading the script first would start a
+/// handler calling a name nobody owns, and every event would be lost.
+pub fn serve(sink: Sender<FocusEvent>) -> Result<Connection, WatcherError> {
+    Builder::session()
+        .map_err(|e| WatcherError::NoSessionBus(e.to_string()))?
+        .name(BUS_NAME)
+        .map_err(|e| WatcherError::NameTaken(e.to_string()))?
+        // `#[zbus::interface]` implements `Interface` for the struct, so the
+        // instance itself is what gets registered — it holds the channel.
+        .serve_at(OBJECT_PATH, Watcher { sink })
+        .map_err(|e| WatcherError::Other(e.to_string()))?
+        .build()
+        .map_err(|e| {
+            // zbus cannot distinguish "name already owned" from other failures
+            // at build time, so the message is what separates them. A name clash
+            // is the common case and deserves the clearer variant.
+            let rendered = e.to_string();
+            if rendered.contains("taken") || rendered.contains("already") {
+                WatcherError::NameTaken(rendered)
+            } else {
+                WatcherError::Other(rendered)
+            }
+        })
+}
+
+/// Drop the bus name, so a restarted daemon does not collide with a stale one.
+///
+/// Called before exiting. Not cosmetic: systemd `Restart=always` can start the
+/// replacement before the old process is fully gone, and losing that race would
+/// leave the new instance wedged on a name its predecessor still holds.
+pub fn release(connection: &Connection) {
+    let _ = connection.release_name(BUS_NAME);
+}