Initial Py_buffer support for Chapel 1D arrays

[jabraham17]

Add initial support for passing Chapel arrays to Python as Py_buffer's and for reading Python objects that support the Py_buffer API.

This enables the following use case

use Python;
proc main() {
  var interp = new Interpreter();
  var mod = new Module(interp, "buffer");
  var doit_numpy = new Function(mod, "doit_numpy");
  var doit_torch = new Function(mod, "doit_torch");
  var doit_numpy_plus_torch = new Function(mod, "doit_numpy_plus_torch");

  var arr = [1, 2, 3, 4, 5];
  var pyArr = new Array(interp, arr);

  writeln("Calling numpy version: ", arr);
  doit_numpy(NoneType, pyArr);
  writeln("Back in Chapel: ", arr);

  arr = [1, 2, 3, 4, 5];
  writeln("Calling torch version: ", arr);
  doit_torch(NoneType, pyArr);
  writeln("Back in Chapel: ", arr);


  arr = [1, 2, 3, 4, 5];
  writeln("Calling numpy+torch version: ", arr);
  doit_numpy_plus_torch(NoneType, pyArr);
  writeln("Back in Chapel: ", arr);
}

import numpy as np
import torch

def doit_numpy(arr):
    x = np.asarray(arr)
    x[:] = 100

def doit_torch(arr):
    t = torch.frombuffer(arr, dtype=torch.int64)
    t[:] = 100

def doit_numpy_plus_torch(arr):
    t = torch.from_numpy(np.asarray(arr))
    t[:] = 100

• st test/library/packages/Python
• st test/library/packages/Python with gasnet

[Reviewed by @lydia-duncan]