Reject malformed base64 payload in RFC 2047 decodeWord

[alhudz]

MimeUtils.decodeWord decodes an RFC 2047 encoded-word reached from a Content-Disposition / Content-Type parameter (filename, name, boundary) via ParameterParser. The base64 branch calls Base64.getMimeDecoder().decode(...), which throws IllegalArgumentException on a malformed final unit. That is not caught by the surrounding catch (IOException), so it escapes the method's declared throws ParseException, UnsupportedEncodingException. The quoted-printable branch throws IOException on the same kind of malformed input and is remapped cleanly, so the two encodings reject bad input inconsistently.

Repro: MimeUtils.decodeText("=?UTF-8?B?A?=") (payload is a single base64 char).
Expected: UnsupportedEncodingException, as for a malformed Q payload.
Actual: IllegalArgumentException ("Last unit does not have enough valid bits").

Widen the existing catch to IOException | IllegalArgumentException so a malformed base64 payload is rejected like a malformed quoted-printable one. Keeping the mapping in decodeWord keeps both encodings within the same declared contract instead of leaking a base64-specific runtime exception. Covered by a new case in MimeUtilityTestCase.

[garydgregory]

Thank you @alhudz, merged 🚀 Please port to 1.x.