This update for afl fixes the following issues:

Updated to 4.21c:

  * afl-fuzz

    - fixed a regression in afl-fuzz that resulted in a 5-10% performace loss
      do a switch from gettimeofday() to clock_gettime() which should be rather
      three times faster. The reason for this is unknown.
    - new queue selection algorithm based on 2 core years of queue data
      analysis. gives a noticable improvement on coverage although the results
      seem counterintuitive :-)
    - added AFL_DISABLE_REDUNDANT for huge queues
    - added `AFL_NO_SYNC` environment variable that does what you think it does
    - fix AFL_PERSISTENT_RECORD
    - run custom_post_process after standard trimming
    - prevent filenames in the queue that have spaces
    - minor fix for FAST schedules
    - more frequent stats update when syncing (todo: check performance impact)
    - now timing of calibration, trimming and syncing is measured seperately,
      thanks to @eqv!
    - -V timing is now accurately the fuzz time (without syncing), before
      long calibration times and syncing could result in now fuzzing being
      made when the time was already run out until then, thanks to @eqv!
    - fix -n uninstrumented mode when ending fuzzing
    - enhanced the ASAN configuration
    - make afl-fuzz use less memory with cmplog and fix a memleak

  * afl-cc:

    - re-enable i386 support that was accidently disabled
    - fixes for LTO and outdated afl-gcc mode for i386
    - fix COMPCOV split compare for old LLVMs
    - disable xml/curl/g_ string transform functions because we do not check
      for null pointers ... TODO
    - ensure shared memory variables are visible in weird build setups
    - compatability to new LLVM 19 changes

  * afl-cmin

    - work with input files that have a space

  * afl-showmap

    - fix memory leak on shmem testcase usage (thanks to @ndrewh)
    - minor fix to collect coverage -C (thanks to @bet4it)

  * Fixed a shmem mmap bug (that rarely came up on MacOS)
  * libtokencap: script generate_libtoken_dict.sh added by @a-shvedov 

Updated to 4.20c:

  + A new forkserver communication model is now introduced. afl-fuzz is
    backward compatible to old compiled targets if they are not built
    for CMPLOG/Redqueen, but new compiled targets will not work with
    old afl-fuzz versions!
  + Recompile all targets that are instrumented for CMPLOG/Redqueen!
  - AFL++ now supports up to 4 billion coverage edges, up from 6 million.
  - New compile option: `make PERFORMANCE=1` - this will enable special
    CPU dependent optimizations that make everything more performant - but
    the binaries will likely won't work on different platforms. Also
    enables a faster hasher if the CPU requirements are met.
  - The persistent record feature (see config.h) was expanded to also
    support replay, thanks to @quarta-qti !
  - afl-fuzz:

    - the new deterministic fuzzing feature is now activated by default,
      deactivate with -z. Parameters -d and -D are ignored.
    - small improvements to CMPLOG/redqueen
    - workround for a bug with MOpt -L when used with -M - in the future
      we will either remove or rewrite MOpt.
    - fix for `-t xxx+` feature
    - -e extension option now saves the queue items, crashes, etc. with the
      extension too
    - fixes for trimmming, correct -V time and reading stats on resume by eqv
      thanks a lot!

  - afl-cc:

    - added collision free caller instrumentation to LTO mode. activate with
      `AFL_LLVM_LTO_CALLER=1`. You can set a max depth to go through single
      block functions with `AFL_LLVM_LTO_CALLER_DEPTH` (default 0)
    - fixes for COMPCOV/LAF and most other modules
    - fix for GCC_PLUGIN cmplog that broke on std::strings

  - afl-whatsup:

    - now also displays current average speed
    - small bugfixes

  - Fixes for aflpp custom mutator and standalone tool
  - Minor edits to afl-persistent-config
  - Prevent temporary files being left behind on aborted afl-whatsup
  - More CPU benchmarks added to benchmark/

Updated to 4.10c:

  - afl-fuzz:

    - default power schedule is now EXPLORE, due a fix in fast schedules
      explore is slightly better now.
    - fixed minor issues in the mutation engine, thanks to @futhewo for
      reporting!
    - better deterministic fuzzing is now available, benchmarks have shown
      to improve fuzzing. Enable with -D. Thanks to @kdsjZh for the PR!
  - afl-cc:

    - large rewrite by @SonicStark which fixes a few corner cases, thanks!
    - LTO mode now requires llvm 12+
    - workaround for ASAN with gcc_plugin mode

  - instrumentation:

    - LLVM 18 support, thanks to @devnexen!
    - Injection (SQL, LDAP, XSS) fuzzing feature now available, see
      `instrumentation/README.injections.md` how to activate/use/expand.
    - compcov/LAF-intel:

      - floating point splitting bug fix by @hexcoder
      - due a bug in LLVM 17 integer splitting is disabled there!
      - when splitting floats was selected, integers were always split as well,
        fixed to require AFL_LLVM_LAF_SPLIT_COMPARES or _ALL as it should

    - dynamic instrumentation filtering for LLVM NATIVE, thanks @Mozilla!
      see utils/dynamic_covfilter/README.md

  - qemu_mode:

    - plugins are now activated by default and a new module is included that
      produces drcov compatible traces for lighthouse/lightkeeper/...
      thanks to @JRomainG to submitting!

  - updated Nyx checkout (fixes a bug) and some QOL
  - updated the custom grammar mutator
  - document afl-cmin does not work on macOS (but afl-cmin.bash does)