Merge pull request #121 in CASA/casa from bugfix/CAS-10150 to master
* commit '2f26b9bf801baac2647e7d8b9361fee603eba467':
CAS-10150 Reduce memory usage by MSMetaData hwen computing subscan props. It is becoming more common for MSes to have several of tens of millions or even hundreds of millions of rows. So, holding several entire columns in memory simultaneously can be very taxing. The somewhat arbitrary limit I've chosen of 10M row blocks seems to put a reasonable cap on memory usage with minimum impact on performance. I may modify it in the future, possibly even letting the client specify it (directly or indirectly) as problematic use cases are identified, but for now, it mitigates the most egregious issues with very large MSes.
