Allow large files to process without running out of memory

[sandbergja]

This patch replaces various occurences where huge MARC files are read into memory as Strings or StringIOs. Instead, it uses:

• TempFiles on disk
• A SAX parser that allows us to read XML files into memory one element at a time, rather than all at once.
• Nokogiri in one case where we were using ReXML

Related to #695

I tried this out on staging, and also the main branch, with 4 files:

branch	peak memory consumption	memory consumption pattern	time
this one	0.4 GB	steady during the whole run	4 minutes 33 seconds
main	1.2 GB	growing at a regular pace during the whole run	4 minutes 7 seconds

For what it's worth, there are 475 files that need to be processed in prod.