PERF-#6728: In the case of narrow dataframes, it is cheaper to convert partitions to numpy in the main process. #14094

Job	Run time
lint (black and isort)	28s
lint (mypy)	3m 35s
execution-filter	6s
lint (flake8)	14s
Test API, headers and no-engine mode	2m 31s
Test BaseOnPython execution, Python 3.9	17m 16s
test-asv-benchmarks	23s
test experimental	15m 3s
test-ubuntu (engine python, python 3.9, group_1)	19m 3s
test-ubuntu (engine ray, python 3.9, group_1)	18m 8s
Test HDK storage format, Python 3.9	38m 25s
test-internals	7m 25s
test-ubuntu (engine ray, python 3.9, group_2)	11m 13s
test-ubuntu (engine ray, python 3.9, group_3)	19m 29s
test-ubuntu-sanity (engine ray, python 3.9)	16m 57s
test-ubuntu (engine ray, python 3.9, group_4)	21m 1s
test-clean-install-ubuntu	1m 32s
test-ubuntu-sanity (engine dask, python 3.9)	17m 22s
test-windows (engine python, python 3.9, group_1)	18m 8s
test-ubuntu-sanity (engine unidist, python 3.9)	16m 48s
test-windows (engine ray, python 3.9, group_1)	18m 14s
test-windows-sanity (engine ray, python 3.9)	15m 33s
test-clean-install-windows	3m 23s
test-windows-sanity (engine dask, python 3.9)	16m 21s
test-windows (engine ray, python 3.9, group_2)	16m 4s
test-windows-sanity (engine unidist, python 3.9)	15m 30s
test-windows (engine ray, python 3.9, group_3)	13m 54s
test-ubuntu (engine unidist mpi, python 3.9)	58m 23s
test-windows (engine ray, python 3.9, group_4)	12m 9s
test (pyarrow, python 3.9)	3m 7s
test-spreadsheet (engine ray, python 3.9)	2m 48s
test-spreadsheet (engine dask, python 3.9)	2m 43s
upload-coverage	25s
	7h 3m 41s

Provide feedback