For dynamic snodes, calling to_numpy() results a whole array padded with zeros, not the actual length

[neozhaoliang]

Let's say we have an empty dynamic snode like this:

import taichi as ti
ti.init(arch=ti.cpu)
x = ti.field(int)
ti.root.dynamic(ti.i, 1024, chunk_size=32).place(x)

@ti.kernel
def test():
    print(x.length())  #0

arr = x.to_numpy()  # array of size 1024

This behavior may cause problems, for example, using arr to draw gui.circle or gui.lines will give false results at the origin.
We should make the returned array have the same size with the actual length of the dynamic snode.

[neozhaoliang]

This might be difficult to solve in the python frontend. Since to_numpy must be called in the python scope, and we cannot know the actual length of a dynamic snode in the python scope. Also, we cannot define a suitable depth for the numpy array if the dynamic snode is a multidimensional array. For a workaround, we should better raise a warning message here.