fix: name the .env path in set_key and unset_key write errors

[jamalkamaladdin]

Fixes #710.

src/dotenv/main.py: an OSError from creating a temporary file in rewrite() now names the .env path.
src/dotenv/cli.py: set and unset print "Error writing env file" and exit with code 2 on an OSError.
tests/test_main.py: 2 cases for set_key with a missing directory and with a read-only directory.
tests/test_cli.py: 2 cases for dotenv set with a missing directory and dotenv unset with a read-only directory.

[theskumar]

Thanks @jamalkamaladdin, this is the right fix. I reproduced #710 on main, including the read-only-directory case for both set_key and unset_key. Setting err.filename and re-raising keeps the original exception type and traceback, and closing source avoids leaking the already-open file.

Why the Windows jobs failed: the code is correct there. The test was the problem: OSError.__str__ quotes the filename with repr(), which doubles every backslash on Windows, so str(dotenv_path) in result.output can never match. The failing log shows the output was otherwise exactly right.

I pushed two commits on top of yours, without changing your commits:

• Merge main: brings in 1.2.4 and Fix a package build deprecation warning #648, so the CHANGELOG entry lands in the right section.
• Follow-up:
  • The two CLI tests now compare the exact (exit_code, output), building the path with {str(dotenv_path)!r}. That works on every OS and matches the literal-value style of the other CLI tests.
  • Adds a CHANGELOG entry crediting you.

I checked the tests by breaking the code on purpose. Every new test catches at least one of these bugs:

Bug introduced on purpose	Caught by
Error still names the temp file	all 4 new tests
set stops catching OSError	test_set_missing_directory
unset stops catching OSError	test_unset_read_only_directory
Exit code 1 instead of 2	both CLI tests

Locally: 264 passed, ruff check/ruff format --check clean, mypy clean on 3.10 and 3.14. I'm waiting for Windows CI on the new head.