Use Int32 for offset arrays

[hombit]

This PR adds an optimizations over list-array offset usage. This makes packer to use the same offset array for all nested columns, optimizing memory usage. This behavior was intended in the initial implementation, but due to the wrong dtype of the offset array it was cast to a new array each time it was used.

[github-actions]

Before [f861313]	After [945e796]	Ratio	Benchmark (Parameter)
31.8±2ms	32.7±3ms	1.03	benchmarks.AssignSingleDfToNestedSeries.time_run
259M	261M	1.01	benchmarks.AssignSingleDfToNestedSeries.peakmem_run
274M	276M	1.01	benchmarks.ReassignHalfOfNestedSeries.peakmem_run
64.6±5ms	65.5±2ms	1.01	benchmarks.ReassignHalfOfNestedSeries.time_run
9.16±0.1ms	9.11±0.07ms	0.99	benchmarks.NestedFrameAddNested.time_run
6.33±0.1ms	6.27±0.06ms	0.99	benchmarks.NestedFrameQuery.time_run
5.26±0.04ms	5.20±0.09ms	0.99	benchmarks.NestedFrameReduce.time_run
89.7M	87.5M	0.98	benchmarks.NestedFrameQuery.peakmem_run
86.4M	84M	0.97	benchmarks.NestedFrameAddNested.peakmem_run
89.7M	86.9M	0.97	benchmarks.NestedFrameReduce.peakmem_run

Click here to view all benchmarks.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.76%. Comparing base (f861313) to head (952c98d).
Report is 63 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #93   +/-   ##
=======================================
  Coverage   98.76%   98.76%           
=======================================
  Files          15       15           
  Lines         893      894    +1     
=======================================
+ Hits          882      883    +1     
  Misses         11       11           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.