[Shanmukh-l33t]

Fixes #6041

Problem

Building bleeding-jumbo with GCC 15 on Ubuntu 26.04 produces 38 warnings: 13 -Wdiscarded-qualifiers, 24 bidi-comment warnings, and one libbz2 #warning (environmental).

Root cause

The -Wdiscarded-qualifiers set is not generic sloppiness in john's code — it needs glibc ≥ 2.40's C23 const-generic strchr/strstr/strrchr overloads (string.h gates them on __GLIBC_USE(ISOC23), which john's own -D_GNU_SOURCE CFLAGS turns on even under -std=gnu99, plus __glibc_const_generic from misc/sys/cdefs.h). On older glibc (this host: 2.39) the same tree builds warning-free, which is why the warnings never showed up before.

Reproduction (RED)

I reproduced the reporter's exact warning set by force-including glibc HEAD's string.h/features.h and __glibc_const_generic, built with GCC 15.2 (the reporter's compiler):

• before the fix: 38 warnings — 13 discarded-qualifiers (formats.c, loader.c, mask.c, sunmd5, hccap2john/wpapcap2john, SIPdump.h, find_the_extra_params, dynamic_expr_normalize, do_compile), 24 bidi-mark warnings in encoding_data.h, 1 libbz2 #warning; matches the issue line-for-line (plus SIPdump.h:335, which the reporter's libpcap-less build skips)
• after the fix: 0 warnings under the same config
• stock gcc-13 build on stock glibc: 0 warnings before and after

Changes

• const on the walkers that only read (formats.c, loader.c, mask.c, sunmd5, hccap2john, wpapcap2john, SIPdump.h, find_the_extra_params)
• dynamic_expr_normalize searches the mutable Buf it just filled rather than re-qualifying a const char *
• do_compile's colon scan reuses the existing const char *cp2
• explicit (char *) casts only where the code genuinely writes through the pointer (link_funcs does *cp++ = 0; comp_optimize's p is a memcpy destination — a const there broke the build and was reverted in favor of the cast)
• encoding_data.h: raw U+200E/U+200F characters inside four comment dumps written as <U+200E>/<U+200F> text (the CP1255/CP1256 tables genuinely map to them and are untouched)

Testing

• reproducing config (GCC 15.2 + glibc HEAD headers): 0 warnings; stock gcc-13 build: 0 warnings
• git show --check: clean (matches CI's whitespace job)
• full ./run/john --test=1: all 439 formats benchmarked, 0 failures
• targeted runtime: --list=formats, --test=1 --format='dynamic=md5($p)', --test=1 --format='raw-md*' (wildcard → wild_cmp/fmt_match), --test=1 --format=SunMD5 (getrounds), mask crack [1][2][3][4] on a known hash (mask_len bracket branch), converter usage runs — all exit 0
• no PR template and no CLA bot in this repository; nothing in the diff changes format behavior — only types, comments, and one scan pointer's qualifier