Create competition dump zip file from stream

When creating a competition dump with big datasets, the datasets must be loaded in to memory to be added to the zip. This is presumably the cause of the Memory exceeded error on dumps. This was fixed by just outputting keys to the datasets used, rather than dumping the datasets. If we dump datasets, should write the zip file from a stream rather than in memory.

Would be nice to add an option to the "create dump" prompt that asks if the user would like to include the dataset files or just the keys. addressed as separate issue: #373