[KennyMcSimpson]

Binary SOAP attachments lose leading and trailing CR/LF bytes because Attachment.content strips them after MultipartDecoder has already removed the MIME boundary framing. This can corrupt downloaded files and invalidate signatures, as reported in #1249 and #1481.

Return binary content unchanged, retaining the existing base64 decoding behavior. Add seven tests built from actual multipart bytes, covering leading/trailing CRLF, an all-CRLF payload, non-text bytes, ordinary content, and empty content. Update three existing fixture expectations to retain the extra CRLF that belongs to their payload; the fixture without that extra CRLF keeps its original expectation.

Validation:

• On unchanged source, five new cases fail and two controls pass.
• Full suites with the project's locked development dependencies on Python 3.10 and 3.13: 483 passed, 17 skipped each.
• Project Ruff lint and format checks, sdist/wheel builds, and Twine checks pass.
• Python's independent email.BytesParser preserves the same bytes for all seven multipart cases. RFC 2046 section 5.1.1 confirms that when two CRLF sequences precede a boundary, the first belongs to the body.

Fixes #1249.
Fixes #1481.

AI assistance: prepared and validated with OpenAI Codex. These are local checks; no human review or successful upstream CI is claimed.