test: assert the consequence, not the setter round-trip - #1925

What this is

A mechanical sweep for pure-echo tests — set a property, assert that same property, no production
call in between — found 15 in the suite:

[Fact]
public void OsLine_Setter_Works()
{
    var vm = NewVm();
    vm.OsLine = "Windows 11 Pro";
    Assert.Equal("Windows 11 Pro", vm.OsLine);   // round-trips a generated setter
}

Every one of these exercises the CommunityToolkit [ObservableProperty] source generator, not this
codebase. They cannot fail on a real defect, and they inflate the test count and the confidence that
number implies. Backlog item #127.

The backlog guessed ~50% false positives. Triaging each one against current source gave a different
and more useful split.

Replaced — the consequence was real and asserted nowhere (5)

Both FilterText properties are wired (OnFilterTextChanged → ApplyFilter), and ApplyFilter
rebuilds the collection the DataGrid binds to. The echo tests never looked at it. A filter wired to
nothing is a defect this project has already shipped — batch 30, five unreachable Services filters —
and the old assertion could not have caught it. Now asserted:

FriendlyEventEntry extreme timestamps — storing MinValue/MaxValue proved a DateTime field
holds a DateTime. The reachable question is what RelativeTime and FullTimestamp do with them,
because RelativeTime subtracts from DateTime.Now, so a future stamp produces a negative span.
Behaviour is pinned as-is (MinValue → em dash, future → "just now") plus an assertion that no negative
duration can reach the user if those branches are ever reordered. Deliberately not "fixed": "just now"
for a clock-corrected event is imprecise but harmless, and far better than printing "in -3d".

Removed — nothing left to assert (9), each verified case by case

• Dashboard OsLine/UptimeLine/CpuPercent/RamPercent _Setter_Works — notification is what a
  binding depends on, and Setter_FiresPropertyChanged already covers it. The two percentages were
  missing from that theory, so they were added to it (parameter widened to object) rather than
  simply dropped.
• Cleanup TempSizeLabel/RecycleBinLabel _CanBeSetDirectly — covered by _DefaultIsScanning
  and PreScan_EventuallyPopulatesLabels.
• Cleanup Progress_AcceptsFullRange — the name implied a 0-100 contract nothing enforces
  (ViewModelBase._progress is unclamped; the ProgressBar clamps visually).
• Performance SelectedPlan_CanBeChanged — default and notification both already covered.
• AppPackage.Status_Transitions_ArePossible, PingTarget.Stats_CanBeUpdated — model echoes with
  no computed members over them.
• DuplicateFile.MinSizeKb_CanBeChanged — its real consequence is minBytes = MinSizeKb * 1024 in
  ScanAsync, and asserting that needs a guard the code does not have: a large typed value overflows
  long and inverts the filter (asking for "≥ 9 quadrillion KB" returns everything). Tracked
  separately so this change stays behaviour-neutral.

Verification

Red proof: 6 mutations, all 3 touched source files restored byte-for-byte. Every replacement goes
red when its claim is broken — which is the whole point, since the tests it replaced could not go red
at all:

No production code changed. All four projects build 0 errors / 0 warnings;
dotnet format --verify-no-changes exit 0; author headers intact on all 9 files; leak scan over all 32
terms gives 0 hits. Net -9 [Fact]/[Theory] attributes (14 removed, 5 added) plus 2 new theory
rows, so the CI count should read 4689. test: — no version bump, no release.