Fix async parser decoding of short ASCII values split across input feeds

[cowtowncoder]

Problem

In NonBlockingByteArrayParser._finishToken(), the MINOR_VALUE_STRING_SHORT_ASCII /
MINOR_VALUE_STRING_SHORT_UNICODE case picks its decoder with:

String text = (_minorState == MINOR_FIELD_NAME_SHORT_ASCII)
        ? _decodeASCIIText(_inputCopy, 0, fullLen)
        : _decodeShortUnicodeText(_inputCopy, 0, fullLen);

MINOR_FIELD_NAME_SHORT_ASCII is 5; the two case labels are 16 and 17. The
condition can therefore never be true, and short ASCII values that span a feed
boundary are always decoded as short Unicode.

For well-formed ASCII content both decoders agree, so this is invisible in normal use.
For content that is not valid ASCII, the same document decodes to different Strings
depending on how the caller chunks the input — a 0xC5 byte inside a short-ASCII token
yields a�cd when fed contiguously but aţd when fed one byte at a time.

The equivalent check for property names (line 233) tests the right constant and is
left unchanged. These are the only two such comparisons in the async parser; the third
(MINOR_HEADER_INLINE, line 396) is reachable and correct. CBOR has no async parser on
2.x, so Smile is the only affected backend.

Fix

Compare against MINOR_VALUE_STRING_SHORT_ASCII.

Test

SimpleStringArrayTest.testShortAsciiValueChunkIndependence writes a short ASCII value,
corrupts one content byte so the two decoders disagree, then asserts that feed sizes of
length+1, 3 and 1 all produce the same String. It fails on the unfixed code with
expected: <a�cd> but was: <aţd>.

Full smile module suite: 270 tests, 0 failures. CI green on JDK 8/17/21.

Note for merge-forward

3.x already carries this same one-token fix via #767, so the merge-forward will conflict
on that line at 3.x (either side is fine to take). 3.1 and 3.2 still have the bug and do
need it. The test will also need the 3.x API names (SmileReadFeature, no IOException)
when it reaches the 3.x side.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BbhNqemoB4csempvUS9n5z