Check for float16 precision support when running translate-* tasks

[gregtatum]

From #934.

One downside of moving [precision and batch sizing] to the [pipeline/translate/decoder.yml] config is that some GPUs on Berlin cluster that we plan to include soon don't support half precision decoding. And if I remember correctly it fails silently and produces bad results on decoding, so it's quite dangerous. So, this setting should be likely propagated through marian args or there should be a detection mechanism in the script.

In the translate script we could check for precision support. Maybe there is some kind of heuristic that could be written to determine the batch size as well based on GPU capacity.