fix: address review feedback and codespell CI failure on #1384

- Preserve valid CP1252 bytes (e.g. 0xe9) in the final decode fallback
  instead of losing them to a UTF-8 replace pass.
- Rename a test fixture from "café.txt" written as caf\xe9.txt, which
  codespell's tokenizer read as the standalone word "caf" and flagged
  as a typo for "calf".
- Add a docstring to test_decode_subprocess_output and a regression
  case mixing a valid CP1252 byte with an undefined one.

Author: claude

dfetch/util/cmdline.py

@@ -50,10 +50,7 @@ def decode_subprocess_output(data: bytes) -> str:
         return data.decode()
     except UnicodeDecodeError:
         pass
-    try:
-        return data.decode(encoding="cp1252")
-    except UnicodeDecodeError:
-        return data.decode(errors="replace")
+    return data.decode(encoding="cp1252", errors="replace")
 
 
 def run_on_cmdline(

tests/test_cmdline.py

@@ -56,7 +56,9 @@ def test_run_on_cmdline(name, cmd, cmd_result, expectation):
         ("utf-8", "café".encode(), "café"),
         ("cp1252 fallback", "café".encode("cp1252"), "café"),
         ("undefined in both codecs is replaced", b"\x81", "�"),
+        ("valid cp1252 byte survives next to an undefined one", b"\xe9\x81", "é�"),
     ],
 )
 def test_decode_subprocess_output(name, data, expected):
+    """Decode UTF-8 and CP1252 output, replacing bytes undecodable in both."""
     assert decode_subprocess_output(data) == expected, name

tests/test_svn_vcs.py

@@ -127,7 +127,7 @@ def test_export_tolerates_non_utf8_output():
     completing.
     """
     with patch("dfetch.vcs.svn.run_on_cmdline") as mock_run:
-        mock_run.return_value.stdout = b"A    caf\xe9.txt\n"
+        mock_run.return_value.stdout = b"A    r\xe9sum\xe9.txt\n"
         SvnRepo.export("svn://example.com/repo", dst="/tmp/out")