MINOR: [C++][Gandiva] cast to unsigned char before ctype calls

[metsw24-max]

Rationale for this change

The Gandiva precompiled string and time helpers feed raw char bytes into isdigit/isxdigit/isalpha/toupper. Where char is signed, any byte above 0x7f (so any multi-byte UTF-8 value coming from untrusted column data) is sign-extended to a negative int, which sits outside the unsigned char/EOF domain those functions are defined for. glibc tolerates it, but a strict ctype implementation indexes its lookup table before the start of the array, which is undefined behaviour. gdv_function_stubs.cc already casts to unsigned char at its call sites; these helpers were missed.

What changes are included in this PR?

Cast each ctype argument to unsigned char at the call site:

• time.cc: castDATE_utf8, castTIMESTAMP_utf8, castTIME_utf8
• string_ops.cc: binary_string, the cast<INT>_varbinary macro, from_hex_utf8, soundex_utf8

The sweep covers both files so no call site is left behind. There is no behaviour change where char is unsigned or the ctype implementation is lenient; it just keeps the argument in range everywhere.

Are these changes tested?

No new test. This is a localised undefined-behaviour fix with no observable change on the platforms CI runs (char is unsigned or the ctype lookup is lenient there), and the existing Gandiva precompiled tests already exercise these functions.

Are there any user-facing changes?

No.

[lriggs]

Looks good.

[metsw24-max]

Rebased onto main to clear the conflict that came in with the recent Gandiva error-message changes (#50110 / #50112). The casts are re-applied unchanged on top of the new error strings, so the diff is the same as what you reviewed. Thanks for the look.

[kou]

Could you use our PR template instead of removing it entirely?

[metsw24-max]

Done, sorry about that. Restored the template and filled in the sections.

[metsw24-max]

Rebased onto main to pick up the tzdata-legacy fix from #50211. The AVX2 failure on the previous run was TestCastTimestampWithTZ (#50173), which came from the conda image moving to Ubuntu 24.04 rather than from this diff, and the JNI/ODBC failures were red on main at the time as well. The casts themselves are unchanged.

[kou]

+1

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 3 benchmarking runs that have been run so far on merge-commit 1828740.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details.