Upgrade to hZCCl2

[JeffeyChou]

This PR brings hZCCL2 (SC'26) into this repository.

What is new in it

• Compressor: introduce 1. block-aligned thread partitioning support; 2. SIMD support(AVX2, AVX2-512); 3. new Outlier-FLE codec; 4. an optional lossless stage (zlib/blosc/zfp)
• Collectives: 1. rebuilt collectives including broadcast, scatter, allreduce; 2. introduce Alltoall collective; 3. Safer datatype and buffer size management, addressing Fix buffer safety, MPI type size correctness, and empty-data edge cases #6 ;
• Benchmarks and jobs: release job scripts associated with artifact.

Fixes

• Every compressed buffer is sized by the codec's worst-case bound, so incompressible or fill-value data can no longer overflow.
• Zero-length inputs are handled by every codec entry point and collective.
• Dropped the library-side MPI_IN_PLACE definition that conflicted with Open MPI.
• Element sizes no longer come from sizeof(MPI_Datatype); collectives accept MPI_FLOAT and return MPI_ERR_TYPE otherwise.

Supersedes #6.