Make uploading large datasets more robust

[tonylampada]

Description

Fixes #163

The upload API may sometimes have some hiccups giving 502 errors with some html in the response.

The pip package handles errors, but doing response.json() when handling an error will itself throw an error, and that can break uploading a big dataset.

This PR makes uploading more robust when using workspace.upload_dataset

Type of change

• Bug fix (non-breaking change which fixes an issue)

How has this change been tested, please provide a testcase or example of how you tested the change?

Uploaded big dataset using a snippet like

testupload.py

#! /Users/tony/.pyenv/shims/python

from roboflow import Roboflow

# export API_URL=https://api.roboflow.one
# export APP_URL=https://app.roboflow.one
# python testupload.py 2>stderr.log | tee stdout.log 

# import roboflow
# roboflow.login(force=True)

rf = Roboflow()
workspace = rf.workspace("tonyprivate2")
dataset_path = '/Users/tony/Downloads/ppe.voc'
workspace.upload_dataset(dataset_path=dataset_path, num_workers=10, project_name="ppe", dataset_format="voc")

Obs: if you separate stdout from stderr with something like python testupload.py 2>stderr.log | tee stdout.logyou get only the bad failures in stderr.

[CLAassistant]

All committers have signed the CLA.

[probicheaux]

lgtm!

[probicheaux]

maybe should be warning or info