Similar to pelias/api#1060, we want to modify this service to call a separate HTTP service for libpostal, rather than loading the 2GB of libpostal data in memory.

This would drastically increase efficiency as we could launch multiple interpolation service instances to scale without paying the cost of loading multiple copies of libpostal.

Since a full planet interpolation dataset is around 40GB, we can take advantage of the Node.js cluster module to allow several separate Node.js processes to use the same copy of the data. This would behave similarly to how we use it with the placeholder service, as mentioned in pelias/placeholder#72.