Fix UnicodeDecodeError reading non-UTF-8 .py module files

[cpruijsen]

extract_metadata in circup/shared.py opened .py files as UTF-8 and let the decode error
propagate, so a single file saved in a Windows ANSI encoding, one with curly quotes in a comment for
instance, crashed the whole scan with UnicodeDecodeError rather than being skipped or read.

This opens those files with errors="replace". The dunder regex only reads the ASCII parts, so
replacing undecodable bytes is enough for cp1252, latin-1 and mac-roman alike, and a file that
previously crashed now yields its __version__ and __repo__.

Files that already decoded as UTF-8 are unaffected.

The new test writes a cp1252 encoded file with a curly quote and asserts the metadata comes back
rather than an exception. The existing test_extract_metadata_python asserts the open() call
signature, so it moves to the new one.

Fixes #222

[mikeysklar]

Would open(path, encoding="utf-8", errors="replace") alone do it? Also, was this AI-assisted? We ask that you mention AI usage when submitting PRs.

[mikeysklar]

I tried the errors="replace" version against a real board (Metro RP2350, CircuitPython 10.4.0-alpha.1) with a cp1252 library on it. list, freeze, install and update all worked, and the board still imported the library fine before and after.

[cpruijsen]

Done in a498363. One thing your suggestion could not see: test_extract_metadata_python asserts the open() call signature, so that assertion moved to include errors="replace". PR description updated to match.

On AI usage: yes, this was AI-assisted. I draft and run the tests with Claude, and review and verify everything before it goes up. Happy to record that in the PR description if you would rather have it there.

[mikeysklar]

Thanks! No need to change this one, but please note AI usage in the PR description going forward.