memory usage can be bad

[andportnoy]

This issue groups three now closed memory-related issues.

1. (improve memory usage with lmer #66)lmer fitting is memory expensive. At least three copies of the data are created when using multiple processes: in the parent process, in the child process, in the R process embedded in the child process. A solution is to use maxtasksperchild to purge R memory regularly. Another solution is to fit channel by channel, which has the same result.

2. (chunk data when running in parallel #67) multiprocessing can't transfer objects > 2 Gb in size. Potential solution is to ensure the chunks are under 2 Gb.

3. (memory #49) Objects can in general be very large. We can del certain attributes (like .data) and conserve a lot of memory.