opa eval command prints errors twice

[tsandall]

For some reason the opa eval subcommand prints errors twice. This seemed to sneak in between v0.10.5 and v0.10.6:

v0.10.6:

$ docker run openpolicyagent/opa:0.10.6 eval '1/0' -f pretty
1/0: eval_internal_error: div: divide by zero
1/0: eval_internal_error: div: divide by zero

v0.10.5:

$ docker run openpolicyagent/opa:0.10.5 eval '1/0' -f pretty
1/0: eval_internal_error: div: divide by zero

This happens regardless of the output format (-f) flag.