fix: require MIME boundaries to start at line boundaries

[audi0417]

The MIME boundary scanners treat any --<boundary> sequence as a delimiter, even when it appears in the middle of a line or is only a prefix of a longer token. This can silently truncate a MIME part without setting is_encoding_problem.

For example, with boundary BND:

visit --BND for details
SECRET

was parsed as:

visit 

The same issue affects the Base64 and quoted-printable MIME decoders, which use the same boundary matching pattern.

Fix

Require MIME boundaries to start at the beginning of a line and validate the bytes following the boundary name. Valid line endings, optional transport padding, and closing boundary markers are accepted; boundary prefixes and body text are left untouched.

Apply the check consistently to multipart parsing, Base64 decoding, quoted-printable decoding, and malformed-message recovery.

Add regression coverage for mid-line boundary-like text, boundary prefixes, valid closing boundaries, and the existing malformed multipart fixtures.

Closes #156

Testing

• cargo test
• cargo clippy --all-targets --all-features -- -D warnings
• git diff --check