IO: memory_map kw in read_csv

[jreback]

not sure of the purpose of memory_map (e.g. whether its needed anymore or not). might be an older option. If it IS current, then needs to be documented.

[gfyoung]

From what I can understand, memory_map differs in that files are mapped onto memory instead of being stored on disk. The code traces back to pandas/src/parser/io.c which includes files like sys/stat.h and sys/mman.h, whose info can be found in the Linux man pages. Thus, I don't even know if this actually applies to Windows for example. It's hard to tell since there is a fall-back if the memory map fails.

Not sure how many people out there use this option (it's not tested too much), and I don't feel comfortable enough writing an explanation about it. Probably someone with a better understanding of memory maps and those header files I mentioned could write something good.

[jreback]

closed by #13381