Decrease memory requirements for /offerings/searchxml

[adamfranco]

Version 1 of the catalog ran with 64M of memory for each process, a pretty light footprint. Most of version 2 works fine with this memory limit, but requests with large result sets like
https://catalog.middlebury.edu/offerings/searchxml/catalog-MCUG/term-202520?enrollable=true&type%255B%255D=genera%253Aoffering-LCT&type%255B%255D=genera%253Aoffering-SEM&type%255B%255D=genera%253Aoffering-LAB&type%255B%255D=genera%253Aoffering-DSC

Fail with 64M and 128M of memory. Bumping to 256M gets it working, but there is room for optimization here. My hunch is that the increase is due to version 2 loading the offering data into an intermediary data structure before freeing the query results, effectively duplicating the total data in-memory, though this could be wrong.

[adamfranco]

It turns out that much of this issue was caused by too many results being included due to double-encoding of entities in the URL as fixed in middlebury/chef#261. There still may be room for improvement, but this is mostly resolved.