Start raster-vision predict job

[lewfish]

This PR adds a method to the Project class that starts a raster-vision predict job on AWS Batch. This job will download the image files associated with the project, and then run object detection on them. To test this PR, you can run the following snippet which will start a job in the R&D account, and will place the predictions for the Oakland project in s3://raster-vision/results/detection/predict/lhf_test/predictions.json. You can then view the output along with the original TIFF file in QGIS. To get the URI for the TIFF file, you can run proj.get_image_source_uris().

from rasterfoundry.api import API
token = ''
api = API(refresh_token=token)
proj = api.projects[0]

proj.start_predict_job(
    's3://raster-vision/results/detection/train/lhf_ships_neg0/train/inference_graph.pb',
    's3://raster-vision/datasets/detection/ships_label_map.pbtxt',
    's3://raster-vision/results/detection/predict/lhf_test/predictions.json')

Closes #29

[lewfish]

The build is failing because the Project class now tries to create a Batch client, and the AWS region isn't set. Any ideas on how to get around this?

[notthatbreezy]

I think we should create the batch client in the function call. Not all Raster Foundry users will have AWS credentials set up or necessarily know about AWS (this is kind of an advanced feature). I don't think this is going to be called enough for a single project that the performance benefit will be noticeable.

This should also fix the build then the client won't be created during tests.

[lewfish]

Good idea.