This PR improves ShareableList's memory usage and performance by consolidating the metadata of each items into a single area in the shared memory block. This allow us to fetch the metadata needed to read and set item using a singlestruct.unpack_from call. Additionally, this MR also removesself._allocated_offsets the offsets can be obtained directly from the shared memory and there is no need to keep a copy in ShareableList. Further more,self._allocated_offsets can has a significant memory footprint when the number of items in the list is large. I am seeing a significant savings in memory footprint and increase in performance in a benchmark where a ShareableList with a list of 10,000,000 zeros is forked by 5 child processes, and each child precess iterate through the list on their own.

Additionally, this PR also include the fix to#26328, and an additional fix where we properly allocate 8 bytes of memory for a string or byte array of size 8. Previously, we would've allocated 16 bytes for a string of size 8 or byte array of size 8.

Here is the resulting memory usage plotted usinghttps://github.com/pythonprofilers/memory_profiler

• Memory consumption of the current ShareableList in multiprocessing
  

• Memory consumption of the ShareableList proposed in this MR
  

https://bugs.python.org/issue46799