fix(ui): stop tests from polluting the live operation-timing.ndjson

DetailsIsland.BindAsync churn (~1900 binds/day, 91-96% cancelled) was not a
UI reactivity bug: OperationTiming.Shared is a static singleton hardcoded to
~/.todo-app/logs/operation-timing.ndjson, and Ui.Tests/Worker.Tests both
construct real DetailsIslandViewModel/TasksIslandViewModel instances that
call Shared.Record directly. Every dotnet test run appended 50-100 lines
straight into the live app's log — pid-burst analysis showed 13 distinct
test-run pids in the ":?" bucket (tests never pass a source) plus a
no-suffix bucket with no pid field at all (pre-dating the pid-per-line
feature). Real user-driven binds that day: 4.

Made Shared settable and added a [ModuleInitializer]-based TestSetup in
both test projects that redirects it to a per-process temp file before any
test runs.

src/ClaudeDo.Ui/CLAUDE.md

@@ -89,7 +89,7 @@ new editor boilerplate there rather than copying it a third time.
 - **IPrimeScheduleApi** — prime-schedule CRUD.
 - **UpdateCheckService** — polls releases; `LastCheckStatus`/`LatestVersion`/`CheckNowAsync` feed the shell's update banner.
 - **InheritanceResolver** — resolves the task → list → global override chain to `(value, source)` for the inherited badges.
-- **OperationTiming** — NDJSON sink (`~/.todo-app/logs/operation-timing.ndjson`, one rolled `.1` at 4 MB) behind every hub invoke and bulk DB path; each line carries `pid` because app restarts interleave in one file. Successful calls under 25 ms are dropped, **failed/cancelled ones always land** — a 1 ms cancelled `BindAsync` is churn signal, not noise. `DetailsIsland.BindAsync:<source>` carries the selection trigger (`TasksIslandViewModel.SelectionSource`, set via `SelectFrom` — never assign `SelectedTask` directly), because the first day of data showed ~1000 binds in 3.5 h, 91% cancelled, with an unexplained trigger.
+- **OperationTiming** — NDJSON sink (`~/.todo-app/logs/operation-timing.ndjson`, one rolled `.1` at 4 MB) behind every hub invoke and bulk DB path; each line carries `pid` because app restarts interleave in one file. Successful calls under 25 ms are dropped, **failed/cancelled ones always land** — a 1 ms cancelled `BindAsync` is churn signal, not noise. `DetailsIsland.BindAsync:<source>` carries the selection trigger (`TasksIslandViewModel.SelectionSource`, set via `SelectFrom` — never assign `SelectedTask` directly). `Shared` is a settable static singleton **on purpose** — `Ui.Tests`/`Worker.Tests` each carry a `TestSetup` module initializer that redirects it to a temp file before any test runs, because both projects construct real ViewModels (`DetailsIslandViewModel`, `TasksIslandViewModel`) that call `Shared.Record` directly; without the redirect a single test run appends 50-100 lines straight into the live app's log (a day of ~1900 `BindAsync` lines, 91-96% cancelled, turned out to be 13 such test runs plus pre-fix-binary artifacts — real user-driven binds that day: 4).
 - **RepoScanner**, **InstallArtifactLocator**/**InstallerLocator**/**WorkerLocator**, **ForegroundHelper** (Win32 foreground before launching a terminal), **FocusClearing**.
 
 ## Converters

src/ClaudeDo.Ui/Services/OperationTiming.cs

@@ -21,7 +21,12 @@ public sealed class OperationTiming
     public static string DefaultPath =>
         Path.Combine(Paths.AppDataRoot(), "logs", "operation-timing.ndjson");
 
-    public static OperationTiming Shared { get; } = new(DefaultPath);
+    // Settable, not just get-only: xUnit constructs real ViewModels (DetailsIslandViewModel,
+    // TasksIslandViewModel, ...) that call Shared.Record directly, so without a way to redirect
+    // it every test run appended its churn straight into the live app's NDJSON file at the real
+    // AppData path — see Ui.Tests' TestSetup module initializer, which points this at a temp file
+    // for the whole test process before any test runs.
+    public static OperationTiming Shared { get; set; } = new(DefaultPath);
 
     private static readonly int ProcessId = Environment.ProcessId;