luajit runs out of memory when training with very big datasets

[dmartinalbo]

The estimation of confidence intervals using tables makes luajit to run out of memory in "EpochSummarizer:summarize" function. Heap size is limited to 2gb.

[bdotgradb]

I created a PR to address this - #30 (but be aware that for large datasets now the sampling time is longer than the epoch time).
Note there are also adjust the following parameters to alleviate this issue without the fix:
--epoch_bootstrap_samples
--num_samples_epoch (drawn from the full training set but the epochs are now smaller which means that you may need to adjust the early stopping and learning rate decay)