Fix build failing with float precision mismatch

[daspecster]

Build is failing due to float precision.
See: https://circleci.com/gh/GoogleCloudPlatform/google-cloud-python/876

Clarification: float (i.e. single) vs. double. Double values are being cast down into float (which is correct, but is also a breaking change). For example

>>> (0.9224355).hex()
'0x1.d849774256bcap-1'>>>fromgoogle.cloud.grpc.vision.v1importimage_annotator_pb2>>>e=image_annotator_pb2.EntityAnnotation(score=0.9224355)
>>>e.score0.9224355220794678>>>e.score.hex()
'0x1.d849780000000p-1'

[dhermes]

@lukesneeringer@geigerj Are you aware of any code paths in google.protobuf that are casting these values from double to float?

[geigerj]

@dhermes I'm not particularly familiar with the inner workings of google.protobuf, but that seems probable given that the source proto declares score as a float, which in the proto wire format is 32-bit (https://developers.google.com/protocol-buffers/docs/encoding).

[dhermes]

@daspecster LGTM. Let's get this merged ASAP. Squash if you'd like

[dhermes]

Whatever issue that caused this seems to have been a regression? I can't can no longer reproduce on Ubuntu 16.04 or OS X 10.11.