Merge pull request TutteInstitute#46 from lmcinnes/master

Stream matrix creation in blocks for lower overhead; if only we could do this with tokenization...