fix: tolerate a malformed encoded-word subject; materialize lengths once

decode_header_value's except tuple missed email.errors.HeaderParseError, so a malformed
base64 encoded-word subject (=?utf-8?B?A?=) crashed the OTP scan instead of falling back
to raw - now caught alongside the other decode errors. retrieve_otp consumed lengths once
per fetched message via set(lengths), so a generator was exhausted after the first message
and a later message's fallback code was missed - list() it once up front like folders.

Signed-off-by: disqualifier <dev@disqualifier.me>

src/aiomail/extract.py

@@ -2,6 +2,7 @@
 import email.message
 import logging
 import re
+from email.errors import HeaderParseError
 from email.header import decode_header, make_header
 from typing import Callable, Iterable, Iterator, Optional, Pattern, Sequence, Union
 
@@ -44,7 +45,7 @@ def decode_header_value(raw: str) -> str:
         return raw
     try:
         return str(make_header(decode_header(raw)))
-    except (UnicodeDecodeError, LookupError, ValueError) as exc:
+    except (UnicodeDecodeError, LookupError, ValueError, HeaderParseError) as exc:
         log.debug("header decode failed (%s): %s", raw, exc)
         return raw
 

src/aiomail/retrieve.py

@@ -75,6 +75,7 @@ async def retrieve_otp(
     forwarded match on From. set `max_age=None` to disable the freshness check.
     """
     folders = list(folders) if folders is not None else list(DEFAULT_FOLDERS)
+    lengths = list(lengths)
     sender_ok = as_predicate(sender)
     subject_ok = as_predicate(subject)
     query = _server_query(sender, subject, match_field)