fix(prs): decode gh/git/claude output as UTF-8, not the Windows cp1252 codec

[LukeTheoJohnson]

Problem

On Windows, graphify prs crashes against any repo that has non-ASCII PR metadata. prs.py reads gh, git, and claude output through subprocess.run(..., text=True) with no explicit encoding=, so stdout is decoded with the locale codec, which is cp1252 on a default Windows install. gh emits UTF-8 JSON whose PR titles and author logins routinely contain non-Latin1 bytes (emoji, or the Persian title in this repo's own PR #1281), and cp1252 has no mapping for bytes such as 0x81.

The failure is indirect. The capture reader thread raises UnicodeDecodeError, subprocess.run swallows that and returns stdout=None, and the caller then dies one line later:

• _gh runs json.loads(None) and raises TypeError
• fetch_pr_files / fetch_worktrees run None.splitlines() and raise AttributeError

Neither is in _gh's except (TimeoutExpired, JSONDecodeError, FileNotFoundError), so the command aborts, and a spurious reader-thread traceback is printed to stderr as well.

Root cause

text=True without encoding= uses locale.getpreferredencoding() for decoding (cp1252 on Windows). CI runs on ubuntu-latest only, where the locale is UTF-8, so this path is never exercised upstream.

Fix

Pass encoding="utf-8", errors="replace" to all five subprocess reads in prs.py. This is the decode-side counterpart of #1505, which applied the same change to the llm.py claude-cli subprocess; prs.py was not touched then.

Reproduction

Windows 11, CPython 3.13, cp1252 locale:

>>> from graphify import prs
>>> prs._gh("pr", "view", "1281", "--repo", "safishamsi/graphify", "--json", "title")
# before: TypeError: the JSON object must be str, bytes or bytearray, not NoneType
# after:  {'title': 'docs: add Persian (فارسی) translation and language selector link'}

Tests

tests/test_prs.py::TestSubprocessOutputEncoding asserts encoding="utf-8" on each of the five calls, mirroring the existing tests/test_charmap_encoding.py. They are red without the fix (the four behavioural asserts fail with assert None == 'utf-8'). A guard test confirms the fixture bytes are genuinely cp1252-undecodable, so the tests exercise the real failure surface. A behavioural decode test can't be cross-platform here, since Linux CI decodes UTF-8 regardless, so the tests assert the contract instead.

Scope

Five subprocess calls in prs.py; no behaviour change on Linux or macOS, where UTF-8 is already what they decode with. Other modules have their own text=True calls but decode ASCII-only output (git hashes, refs), so I left them out to keep this focused. Happy to harden those in a follow-up if you'd like.

[safishamsi]

Landed on v8 (ships in 0.9.19) in a86666a, cherry-picked to preserve your authorship. Thanks @luke J — all gh/git/claude subprocess reads in prs.py now decode as UTF-8 with errors=replace (matching #1505), fixing the Windows cp1252 mojibake/crash on emoji and non-ASCII PR titles/diffs.