Additional fixes

[dehall]

1. Don't consider missing addresses at all in household inference. Originally the logic would strip these out at the end but one data owner revealed this could cause performance issues, so now strip them out before indexing. This means these rows will not be considered at all, and will be their own "household". At some point we should probably also consider the option of not assigning individuals a household at all.
2. Remove textdistance "extras" from requirements.txt since some users had trouble with this on windows. A note will be added to the installation instructions on the wiki that installing extras is optional
3. Fix an issue with counting addresses in data_analysis.py. The issue is related to how the code combines address_street and address_detail, since it gets called from both data_analysis.py and extract.py.
4. Count blank strings as "missing" in data_analysis
5. Wrap some values in max(int(x), 1) in case x is a fraction 0 < x < 1

[dehall]

I thought I merged this months ago