fix(app): keep the exit save overlay stable while disks save in parallel

[coldhighsun]

Summary

When several RAM disks are saved in parallel on exit, each progress report overwrote the overlay with its own disk's name and byte counts, so the status text and progress bar flickered between disks. Reports could also be delivered out of order, nudging the bar backwards.

The overlay now shows an aggregate instead of whichever disk reported last.

Changes

• ExitSaveProgressTracker (new, thread-safe): folds the per-disk reports into one monotonic, size-weighted state with a fixed byte total known before the first report. The bar and the byte count share a 99% ceiling until the save is completed, so an untracked disk that is still saving cannot be hidden behind a full bar.
• Overlay texts: "Saving N disk images (k/N done)" for several disks, the disk's own name for one, "Finishing up..." once every disk is done, and "M of N disk images could not be saved" when the save ended with failures. Strings are added in both languages.
• Refresh path: the save threads only update the tracker. The screen is refreshed through the new CoalescingSignal at Background priority, so a burst of reports cannot flood the dispatcher or delay input.
• Failures: SaveFailed marks a disk as failed only once its own final save has begun, so an earlier periodic auto-save failure does not mark a disk whose final save then succeeds. A failed disk is never shown as finished.
• MountManager.Dispose(Action<RamDisk, double>): the callback now reports only the disk and its own fraction. The overall fraction, the used-byte count and DisposeProgressAggregator (with its tests) are removed, since the app aggregates by size itself.
• RamDisk.WillSaveOnDispose (backed by RamDiskSaveDecisions): tells the app which disks will really be written, so clean disks are not counted as "disk images".
• Shutdown robustness: collecting the sizes and finishing the display are isolated teardown steps, so a failure there cannot skip the exit save.
• CLAUDE.md documents the exit-save progress pipeline and its ordering rules.

Notes

• MountManager.Dispose(Action<RamDisk, double, double, ulong>) is replaced by a two-argument overload. The app is the only caller.
• The wiring in App.ShutdownAsync (CollectExitSaveSizes, the SaveFailed hookup, BeginExitSave before Dispose) is not unit-tested because it needs a real RamDisk (WinFsp driver) or the WPF App.
• SettingsStoreTests.Update_ConcurrentUpdatesOfDifferentFields_KeepsEveryChange failed intermittently on some runs. It is unrelated to this change and passes on rerun.

Test plan

• dotnet build ManagedDrive.slnx (no warnings)
• dotnet test tests/ManagedDrive.Tests (1859 tests; new tests for the tracker, CoalescingSignal, the view model texts and the localization keys)
• Manual: mount 3 disks with data, exit, and check the text is stable and the bar and bytes only increase
• Manual: exit with unchanged disks and check the overlay keeps "Saving disk images, please wait..."

[codecov]

Codecov Report

❌ Patch coverage is 66.66667% with 40 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/ManagedDrive.App/App.xaml.cs	0.00%	29 Missing ⚠️
src/ManagedDrive.Core/Mounting/MountManager.cs	0.00%	6 Missing ⚠️
...p/ViewModels/MainViewModel/MainViewModel.Status.cs	91.30%	1 Missing and 1 partial ⚠️
src/ManagedDrive.Core/Mounting/RamDisk.cs	0.00%	2 Missing ⚠️
...nagedDrive.App/Services/ExitSaveProgressTracker.cs	98.14%	0 Missing and 1 partial ⚠️

📢 Thoughts on this report? Let us know!