fix: a downloaded file with broken pixel data is skipped instead of ending the download run

Importer._import_media and Importer.attach_in_place each carried their own
copy of the still-image filters (verify, minimum size, transparency, single
colour) and of the phash step, and the copies had drifted: attach_in_place's
transparency check had no guard for an OSError from decoding truncated pixel
data, so that error escaped into DownloadService's per-file loop and ended the
run with the remaining files unattached; its phash step swallowed every error
and imported the broken file without a hash.

Both paths now call _screen_still and _phash_of, which turn a decode error
into an invalid_image skip.

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

tests/test_importer.py

@@ -216,6 +216,28 @@ def test_pil_load_oserror_in_transparency_check_skips_not_raises(
     assert "transparency check" in (result.error or "")
 
 
+def test_downloaded_file_with_broken_pixels_skips_not_raises(
+    importer, import_layout, monkeypatch,
+):
+    """A downloaded file goes through the same filters as an imported one, so
+    a decode error there is a skip; raised, it would end the download run
+    before the rest of its files were attached."""
+    _, images_root = import_layout
+    path = images_root / "Bob" / "corrupt.png"
+    _make_png_rgba(path, (100, 100), alpha=128)
+    importer.settings.skip_transparent = True
+    importer.settings.transparency_threshold = 0.5
+
+    def _boom(_self, _src):
+        raise OSError("broken data stream when reading image file")
+    monkeypatch.setattr(type(importer), "_transparency_pct", _boom)
+
+    result = importer.attach_in_place(path)
+    assert result.status == "skipped"
+    assert result.skip_reason == SkipReason.invalid_image
+    assert "transparency check" in (result.error or "")
+
+
 def test_pil_load_oserror_in_phash_compute_skips_not_raises(
     importer, import_layout, monkeypatch,
 ):