detect_text_encoding has no EBCDIC branch, so EBCDIC text is reported as binary #3507

Summary

libmagic recognizes EBCDIC text by translating the buffer to ASCII through a lookup table and
re-testing it. PolyFile's detect_text_encoding has no EBCDIC branch, so EBCDIC text is classified
as binary and falls through to application/octet-stream.

Reproducer

$ printf '\x7fELF' > elf.bin
$ ./file/src/file -m tests/magic.mgc -i --keep-going elf.bin
elf.bin: text/plain; charset=ebcdic
$ ./file/src/file -m tests/magic.mgc --keep-going elf.bin
elf.bin: ELF\012- , International EBCDIC text, with no line terminators

from polyfile.magic import MAGIC_DEFS, MagicMatcher, MatchContext, detect_text_encoding

matcher = MagicMatcher.parse(*MAGIC_DEFS)
print(detect_text_encoding(open("elf.bin", "rb").read()))
print({mt for m in matcher.match(MatchContext.load("elf.bin", only_match_mime=True))
       for mt in m.mimetypes})

None
{'application/octet-stream'}

Expected an EBCDIC encoding name and {'text/plain'}.

Cause

file_encoding tries EBCDIC after the ASCII, UTF-8 and UCS families fail
(file/src/encoding.c:159-169):

from_ebcdic(buf, nbytes, nbuf);

if (looks_ascii(nbuf, nbytes, *ubuf, ulen)) {
        DPRINTF(("ebcdic %" SIZE_T_FORMAT "u\n", *ulen));
        *code = "EBCDIC";
        *code_mime = "ebcdic";
        ...
} else if (looks_latin1(nbuf, nbytes, *ubuf, ulen)) {
        DPRINTF(("ebcdic/international %" SIZE_T_FORMAT "u\n", *ulen));
        *code = "International EBCDIC";
        *code_mime = "ebcdic";

from_ebcdic (file/src/encoding.c:657) is a byte-for-byte table translation. Two tables back it:
ebcdic_to_ascii (file/src/encoding.c:599) and ebcdic_1047_to_8859 (file/src/encoding.c:633),
256 bytes each. The comment at encoding.c:583-597 notes that the five documented EBCDIC variants
agree on the characters that matter for this test, so one table is enough to identify EBCDIC text.

detect_text_encoding (polyfile/magic.py:3031) implements the ASCII, UTF-8, UCS and 8-bit
branches and stops there, returning None. MagicMatcher.match then takes the OctetStreamTest
branch (polyfile/magic.py:3453-3456).

Note the ordering constraint: EBCDIC is tried only after the other families fail, because most
EBCDIC buffers are also valid 8-bit buffers. Inserting the branch anywhere earlier would reclassify
ordinary Latin-1 text.

Impact

pocs-master/mocks/elf.bin in the corkami corpus, currently skipped by KNOWN_BAD_FILES in
tests/test_corkami.py. Its skip comment claimed PolyFile was the more accurate of the two, which
no longer holds after #3468.

Worth weighing before picking this up: the payload is two 256-byte tables plus a translation pass,
and EBCDIC input is rare outside mainframe data. It is a real fidelity gap against file rather
than an urgent one, and the corpus entry can stay skipped until someone wants the coverage.

Independent of #3488, which changes when the plain-text result is emitted rather than whether these
bytes count as text. Related to #3506, the missing NUL-trimming step in the same function.