fix: the single-color filter no longer takes line art for a blank image (#4483)

Five of Todding's Discord doodles (pencil lines on white, 3000px) were skipped
on import as "single color", so their posts showed text and no image. The
predicate sampled a 64px BILINEAR thumbnail, which blends thin strokes into
the paper, and 0.95 "one color" is below how white a doodle is.

- Sample 256x256 by NEAREST, so each sample is a real pixel.
- Default threshold 0.995: blank means essentially blank. Migration 0116 moves
  a stored 0.95 (the old default) with it; the settings slider now spans
  0.9-1 in 0.005 steps so the value is reachable.
- The Cleanup audit shares the predicate, so it stops flagging sketches too.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

backend/app/models/import_settings.py

@@ -39,7 +39,7 @@ class ImportSettings(Base):
     transparency_threshold: Mapped[float] = mapped_column(Float, nullable=False, default=0.9, server_default="0.9")
 
     skip_single_color: Mapped[bool] = mapped_column(Boolean, nullable=False, default=False, server_default="false")
-    single_color_threshold: Mapped[float] = mapped_column(Float, nullable=False, default=0.95, server_default="0.95")
+    single_color_threshold: Mapped[float] = mapped_column(Float, nullable=False, default=0.995, server_default="0.995")
     single_color_tolerance: Mapped[int] = mapped_column(Integer, nullable=False, default=30, server_default="30")
 
     # Hamming distance over a 256-bit pHash (utils.phash, hash_size=16). The

backend/app/services/audits/single_color.py

@@ -4,11 +4,19 @@ predicate for BOTH surfaces: FC-Cleanup's retroactive audit and — since
 2026-07-02 — the import-side filter (Importer._single_color_hit /
 SkipReason.single_color), so what the audit flags and what the import
 skips can never disagree.
+
+It is meant to catch the blank: a placeholder, an error tile, a solid fill.
+Line art is the case it must not catch (#4483): a pencil doodle on white can be
+well over 95% white even at full size, and a smoothing downsample blends its strokes
+into the paper until it measures as blank. So the sample is taken by NEAREST
+(each sampled pixel is a real pixel, strokes keep their contrast), at 256px
+so thin strokes are still hit, and the default threshold is near-total
+(0.995) — a few percent of ink is a drawing, not an empty image.
 """
 
 from PIL import Image
 
-_THUMB_SIZE = (64, 64)
+_THUMB_SIZE = (256, 256)
 
 
 def evaluate(
@@ -20,7 +28,8 @@ def evaluate(
     """True iff the fraction of pixels within `tolerance` (Euclidean RGB
     distance) of the dominant color exceeds `threshold`.
 
-    Downsamples to 64x64 for speed (~4ms regardless of source size).
+    Samples 256x256 by NEAREST (see the module docstring for why not a
+    smoothing resample).
     Alpha channels are stripped; only RGB is considered. Animated images
     use frame 0 (PIL's default after Image.open without seek).
     """
@@ -30,7 +39,7 @@ def evaluate(
     elif im.mode not in ("RGB", "L"):
         im = im.convert("RGB")
     if im.size != _THUMB_SIZE:
-        im = im.resize(_THUMB_SIZE, Image.Resampling.BILINEAR)
+        im = im.resize(_THUMB_SIZE, Image.Resampling.NEAREST)
     pixels = list(im.getdata())
     if not pixels:
         return False