Fix RFC 2231 continuation parameter decoding (multi-octet split + charset inheritance)

[gaoflow]

RFC 2231 §3/§4.1 lets a long parameter value be split into numbered continuation
sections. The charset'lang' prefix appears only on section 0, later encoded sections
carry raw percent-encoded octets in that same charset, and a multi-octet character may
legally be split across sections. The correct decode is: percent-decode each encoded
section to octets, concatenate all octets, then charset-decode once.

parse_content_type instead decodes each section independently — percent-decode and
charset-decode the section's bytes on their own — and concatenates the resulting strings,
resetting the charset between sections. Two things break.

1. A multi-octet character split across sections decodes each fragment alone, so each
becomes U+FFFD:

Content-Type: application/x-stuff; name*0*=UTF-8''%E2%98; name*1*=%83.txt

got       "\u{FFFD}\u{FFFD}.txt"
expected  "☃.txt"

%E2%98 and %83 are the three UTF-8 bytes of U+2603. Both Go's mime.ParseMediaType
and Python's email (collapse_rfc2231_value) decode this header to ☃.txt.

2. The charset given only on section 0 is not inherited by later standard-form
sections, so their non-ASCII octets fall back to UTF-8:

Content-Type: application/x-stuff; name*0*=ISO-8859-1''%E9; name*1*=%E8.txt

got       "é\u{FFFD}.txt"
expected  "éè.txt"     (Python email agrees)

Fix

Keep each encoded section's percent-decoded octets, resolve the charset from section 0
(still honoring an explicit per-section charset' prefix, so inputs that repeat it decode
as before), concatenate the octets across sections, then charset-decode once. Literal
(non-*) sections are appended verbatim.

New content_type.json cases cover the UTF-8 split above, the section-0-only charset,
out-of-order sections (*1* before *0*, charset still taken from index 0), and an
encoded section followed by a literal one in a non-UTF-8 charset. The existing
continuation tests stay byte-identical — including the non-standard
filename*0*=iso-8859-1'es'… case that repeats the charset on every section, the
title*0*=…; title*2="…" mixed encoded/literal case, and the out-of-order
name*1=…; name*0=…; name*2*=… integration test.

cargo test and cargo test --features full_encoding both pass.