Fix NumPy 2 overflow and quadratic point selection on large stacks

[A220N1]

Two independent problems met while processing Sentinel-1 stacks of a few million
second-order points.

prog_bar.update(..., every=np.int16(...)) raises OverflowError: Python integer 32768 out of bounds for int16 under NumPy 2, at the very end of step 4 and during
triangulation = the whole run is lost.

The three point selection loops (objects.removePoints,
processing.runDensificationTimeAndSpace twice) are quadratic in the number of
points and never complete at that size. np.isin gives the same mask, duplicates
included.

Checked on a 55-date Sentinel-1 stack: steps 0 to 4 run twice, with and without the
vectorization, give 49 140 points, the same identifiers and a maximum difference of
0 over the 62 numeric columns of the export.

No test is added: both changes are internal and the existing suite covers them
through test_processing.

[A220N1]

A third problem, same family. gstools evaluates a kriging model in a single
block by default (chunk_size=None), so filtering.py allocates one
(conditioning points x evaluated points) float64 array, with a few thousand
first-order points interpolated onto a few million second-order ones, that is
hundreds of gigabytes and the worker is killed.

Evaluating in blocks of at most 128 MB is bit-identical and costs nothing.

[Andreas-Piter]

Dear @A220N1,
Thanks for pointing to these issues in the implementation and for providing a fix. I have tested it and it can be merged.
Thank you so much for contributing to our community software! If you have any further comments, suggestions or feedback, please send me an email or add an issue here on github.
Best, Andreas