Quote keys containing '=' when writing (fixes #273)

[gaoflow]

Summary

Fixes #273. A config key containing = is written unquoted, so on the next parse the first = is read as the key/value divider — silently splitting the key and corrupting the value:

from configobj import ConfigObj
import io

co = ConfigObj()
co['a = b'] = 'val'
buf = io.BytesIO(); co.write(buf)          # writes:  a = b = val
buf.seek(0)
dict(ConfigObj(buf))                        # -> {'a': 'b = val'}   (corrupted)

The parser already accepts the correct quoted form ("a = b" = val) — only the writer fails to emit the quotes. The issue #273 example "x <a='b'>" is worse: it produces a file that raises ParseError on the next read.

Cause

_quote() is shared by keys, values, list elements and section names. Its "no quoting needed" branch checks for leading/trailing whitespace-plus chars and ,, but not =. That is correct for values, list elements and section names — = is legal unquoted there — but not for a key, where the first = on the line is the divider. _write_line quoted the key with no key-context flag.

Fix

Add an is_key flag to _quote(). When it is set and the key contains =, force a quoted form via _get_single_quote() (which already picks "/' correctly, even when the key itself contains a quote char). _write_line passes is_key=True for the keyword; values, list elements and section markers are unchanged (is_key=False).

The guard sits ahead of both no-quote paths, so it also covers list_values=False (where the same corruption occurred).

Verification

• New regression test test_key_with_equals_is_quoted asserts the written form ("a = b" = val) and that it round-trips back to {'a = b': 'val'}. It fails before the fix and passes after.
• Confirmed by hand: the minimal repro, the list_values=False path, issue removed quotation marks if writing keys containing equal sign and "less than" sign #273's "x <a='b'>" example, and nested keys all round-trip; section names containing = stay unquoted ([sec = x]) and round-trip; normal keys and values are written exactly as before.
• Full suite: 80 passed (was 79). flake8 --max-line-length=132 introduces zero new warnings.

This pull request was prepared with the assistance of AI, under my direction and review.