[WIP] Write data to compact regions through destination "filling"

[tbagrel1]

This PR provides a way to fill compact regions (GHC.Compact) through destination filling, with mutations under the hood.

It aims to reduce the number of allocations needed to copy data in a compact region for specific workloads such as parsers or deserializers.

[tbagrel1]

Here's a performance report at commit c86e68f:

Benchmark memory: RUNNING...
All                      
  compact region allocs  
    parser without dest: OK (0.43s)
      203  μs ±  16 μs, 1.1 MB allocated,  18 KB copied,  10 MB peak memory
    parser using dest:   OK (0.13s)
      253  μs ±  21 μs, 1.4 MB allocated,  10 KB copied,  10 MB peak memory

memory_using_dest_c86e68fca2a3f1ce4ea5aa18c31e336059d1ccbd.txt
memory_without_dest_c86e68fca2a3f1ce4ea5aa18c31e336059d1ccbd.txt

[tbagrel1]

I'm reviving this PR after one year.

The new implementation works with a custom GHC version, that adds 3 new primitives to efficiently allocate inside compact regions without copying.

It is based on my JFLA article https://inria.hal.science/hal-04406360/document and HIW presentation https://www.youtube.com/live/uMurx1a6Zck?si=kNR2Qsj-6hVGdlP2&t=10579