Handle objects larger than available disk space gracefully in the Client FileCache

[tuulos]

Currently the Client FileCache tries to write each object on disk before loading it to memory. This is problematic when the available disk space is less than the available memory, which can happen e.g. on the default AMIs for AWS Batch:

https://gitter.im/metaflow_org/community?at=5e5dfb2dd045e258250e782f

In cases like this, we should just skip the cache.