Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@PhotoLocator/BitmapOperations/AstroStretchOperation.cs`:
- Around line 46-48: The computed sampled width can evaluate to 0 for very
tall/narrow images causing an empty FloatBitmap and downstream divide-by-zero in
Mean(); update the resize code in AstroStretchOperation (around
ConvertToGrayscaleOperation.ConvertToGrayscale, FloatBitmap creation and
BilinearResizeOperation.ApplyToPlaneParallel) to compute a sampledWidth and
clamp it to a minimum of 1 (e.g. sampledWidth = Math.Max(1, calculatedWidth)
using an appropriate rounding), then construct the FloatBitmap with that clamped
width and SampleHeight before calling
BilinearResizeOperation.ApplyToPlaneParallel.

In `@PhotoLocator/LocalContrastViewModel.cs`:
- Around line 100-106: The IsAstroModeEnabled setter only raises PropertyChanged
for itself via SetProperty(ref field, value) but neglects to notify the
dependent IsLocalAdjustmentsLabelVisible and to refresh the preview pipeline;
update the setter in IsAstroModeEnabled so that after a successful SetProperty
it also calls OnPropertyChanged(nameof(IsLocalAdjustmentsLabelVisible)) (or
SetProperty overload that notifies both) and then triggers the pipeline/preview
refresh by calling the existing refresh method used elsewhere (e.g.,
SchedulePipelineRefresh(), RefreshPreview() or equivalent) so toggling astro
mode immediately updates the label and preview.

In `@PhotoLocatorTest/BitmapOperations/AstroStretchOperationTest.cs`:
- Line 32: The test AstroStretchOperationTest currently asserts
Assert.AreEqual(0.0026545193517195226, op.DstBitmap.Mean(), 0.000001) which is
too tight for JPEG/floating-point variations; change the assertion to use a
looser tolerance (for example 1e-4 or 1e-5) or assert the mean falls within an
explicit range (e.g., Assert.IsTrue(Math.Abs(op.DstBitmap.Mean() -
0.0026545193517195226) < 1e-4) or Assert.That(op.DstBitmap.Mean(),
Is.InRange(expected - delta, expected + delta))) so the check on
op.DstBitmap.Mean() is robust across environments.

---

Outside diff comments:
In `@PhotoLocator/LocalContrastViewModel.cs`:
- Around line 573-589: The no-op fast path (IsNoOperation) currently ignores
astro-related settings so ApplyOperations can incorrectly skip astro-only edits;
update the IsNoOperation getter to also require astro mode and astro parameters
be at their defaults—e.g., include checks on IsAstroModeEnabled (or
_astroOperation.IsAstroModeEnabled), AstroStretch == DefaultAstroStretch,
BackgroundRemovalSmooth == DefaultBackgroundRemovalSmooth and BlackPoint ==
DefaultBlackPoint (or the corresponding properties/fields) so that any
non-default astro setting prevents returning the original source and forces the
astro pipeline (ApplyAstroStretchOperation and related steps) to run.

---

Nitpick comments:
In `@PhotoLocator/JpegTransformCommands.cs`:
- Line 106: The code in JpegTransformCommands.cs constructs
LocalContrastViewModel with IsAstroModeEnabled = o as string == "Astro", which
hardcodes the "Astro" command; replace that literal with a shared constant (e.g.
add a public const string Astro = "Astro" in a Commands or Constants class and
use Commands.Astro here) and update any other places (including XAML bindings
that reference the literal) to reference the same constant (use x:Static in XAML
if applicable) so renames remain safe; change the equality check in the
JpegTransformCommands method and any other comparisons to use the new constant
instead of the string literal.