optimise (search-index) memory usage

[blahah]

See https://github.com/fergiemcdowall/search-index/issues/261 and https://github.com/fergiemcdowall/search-index-adder/issues/2

Some of the issues are addressed by the NLP pipeline which greatly reduces redundancy in the index. However, memory usage is still high.

A test case using 10,000 publication metadata records:

npm install --global yuno
wget https://git.io/vw48e
wget https://git.io/vw48L
yuno create --opts eupmc_db_opts.json memory_test eupmc_lite_10k.json

[eklem]

The new search-index streaming API does wonders here. Where I maxed out with 16 GB of memory before, I now manage with 1 GB of memory.

[blahah]

yes! see #6 - this is why we sponsored the streaming interface in search-index :)