Look into eliminating the --optimize_for_large_inputs flag

[arostamianfar]

We can likely detect whether the input is "large" at initialization and automate setting this flag. We already try to do some of this (e.g. https://github.com/googlegenomics/gcp-variant-transforms/blob/master/gcp_variant_transforms/vcf_to_bq_common.py#L69), but it needs to be smarter and also take size of the files into account.

This can make the user experience nicer as they don't need to worry about passing an additional flag.

Note: we should make sure that the initialization itself does not take too long for very large inputs (e.g. add a cap to the number of files being processed).