Limit vtkDelimitedTextReader memory consumption

vtkDelimitedTextReader memory footprint is now much lower during computation. (i.e. inside the RequestData, called from Update)

It used to read all the data as string array and then have a post-process pass to create numerical (int or double) arrays. Thus, at some point the data was duplicated as string and numeric.

Instead, check on-the-fly if conversion is possible, to minimize the string allocation (and duplication)

Kind of benchmark

Loading a CSV file of 100k lines, with 9 numerical columns (3 int and 9 double) with this python script

import argparse
import vtk

parser = argparse.ArgumentParser()
parser.add_argument('-f', '--filepath')
args = parser.parse_args()
filepath = args.filepath

reader = vtk.vtkDelimitedTextReader()
reader.SetFileName(filepath)
reader.SetHaveHeaders(True)
reader.DetectNumericColumnsOn()

reader.Update()

Running through heaptrack gives the following informations:

version	peak memory	peak contribution	calls to allocation	duration
master	58.5 MB	41.9 MB from vtkStringArray	1,042,111	0.873 s
this topic	17.9 MB	6.3 MB from vtkBuffer	39,784	0.344 s