Reject a key that set_key cannot write faithfully

[lenamonj]

set_key validates quote_mode and quotes the value it writes, but nothing validates the key. The key is interpolated straight into the line, so a key the parser reads back differently is silently stored as a different variable, while the call returns (True, key, value) to report success.

>>> from dotenv import set_key, dotenv_values
>>> set_key(".env", "a=b", "x")
(True, 'a=b', 'x')
>>> open(".env").read()
"a=b='x'\n"
>>> dotenv_values(".env")
{'a': "b='x'"}

The caller asked for key a=b and got key a holding b='x'.

Other shapes are lost outright rather than altered. "a b", " a", "a\tb" and "" write lines the parser rejects, so the key silently does not exist afterwards; "a\nb" writes a second line, so a spurious extra key appears. Each returns True.

All of it is reachable from the CLI:

$ dotenv -f .env set "a=b" x
a=b=x
$ echo $?
0
$ dotenv -f .env list
a=b='x'

The change

Validate by round trip rather than by restating the parser's rules in a second place: write a probe line with a trivial value and check that the parser hands the same key back. If the parser and set_key ever disagree about what a key is, this check follows the parser automatically.

Keys that work today are unaffected. The added test pins that explicitly for a, A_B1, a.b, MY-KEY, a'b, a$b and a non-ASCII key.

ValueError matches the existing behaviour for a bad quote_mode a few lines above.

Scope is deliberately the key only. The value side of quote_mode="never" is #218, which you closed after #330, and this change does not touch it.

Verification

Against main at 751f8c14:

• The two added tests, applied alone to unmodified main: 8 failed, 7 passed. The 8 failures are exactly the 8 rejection cases; the 7 passes are the control keys, which already round-trip correctly.
• With the change: 235 passed, 1 skipped (the suite was 220 passed, 1 skipped before the 15 added cases).
• ruff check ., ruff format --check src tests and mypy src all clean.