No OSX support for multithreading

[mklarqvist]

What
Our implementation makes use of Multiprocessing.queue and more specifically qsize that is not implemented in OSX.
https://docs.python.org/2/library/multiprocessing.html#multiprocessing.Queue.qsize. Attempting to run on OSX will result in a fatal error and then our manually spawned threads will never join resulting in an infinite wait.

Minimally reproducible code:

import ml4cvd
from ml4cvd.arguments import tensormap_lookup
from ml4cvd.tensor_generators import TensorGenerator
import glob

# files = glob.glob("/tf/google-ecg-rest-38k-tensors/2020-03-14/*.hd5")
files = glob.glob("2020-03-14/*.hd5")
batch_size = 100
tgen = TensorGenerator(
    batch_size = batch_size, 
    input_maps = [tensormap_lookup('ukb.ecg.ecg_rest')], 
    output_maps = [tensormap_lookup('ukb.demographics.age_0')], 
    num_workers = 8, 
    paths = files, 
    cache_size = 0
)
iterator = iter(tgen)

(internal code spawns workers if this is the first iteration)

>>> next(iterator)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/miniconda3/envs/ml4cvd/lib/python3.6/site-packages/ml4cvd/tensor_generators.py", line 164, in __next__
    if self.stats_q.qsize() == self.num_workers:
  File "/opt/miniconda3/envs/ml4cvd/lib/python3.6/multiprocessing/queues.py", line 117, in qsize
    return self._maxsize - self._sem._semlock._get_value()
NotImplementedError

[mklarqvist]

This is partially related to issue #352

[mklarqvist]

This will be addressed when migrating to tf.data.Dataset-based approaches