download function should avoid race-condition

[szha]

Currently the download function would directly write to target file path, which causes file integrity problem when multiple processes use the same logic to write to the same file.
https://github.com/apache/incubator-mxnet/blob/master/python/mxnet/gluon/utils.py#L223-L226

Alternative is to write to a temporary file first, and only move to target path upon download completion after verifying file integrity. Temporary file should be cleaned up at the end. This should help solve issues like #11616

[nswamy]

You could use tempfile.TemporaryFile or tempfile.NamedTemporaryFile that automatically cleans up after its closed