32 vs 64 float incompatibility

[alexhutko]

I tried to upload a float 32 to a metric and it barfed until I converted it to float 64. This is quite an inconvenient onus to place on the user, particularly since there is no place in in the metric object to clarify this. Can this be remedied somehow? The error message offers some hint, but isn't particularly graceful:

AttributeError: 'numpy.float32' object has not attribute 'swagger_types'

Lemme know if you want me to test something out on staging if it gets there.