But if you add this flag, you will loose valuable tuned-table information perhaps ran on a system with a representative dataset. 

<AverageFieldSize>9.62</AverageFieldSize>

<Selectivity>0.7246%</Selectivity>

<BlockCount>-12</BlockCount>

OutlierSelectivity also is nice but contains "real data" and should be avoided. On our end we plan to run a "linting" script after exportation but before commit to keep valuable stats but omit the data itself (such as the one in the histogram). 

Does this seems like a good way to do it ? 

That would work and is a clean solution. This is what I expected from the the framework.

I appreciate the suggestion. However this would only work for a JSON Export and not a JSON Import.