improve pipeline garbage collection capabilities

[bstabler]

The current full example run (see #194 for example) is running into memory issues. @toliwaga guesses the intermediate tables for logsum samples (which are enormous) need to get flushed – currently they are taking up memory and space in the pipeline because the three components of the logsum models are separate steps and so the tables are saved as orca tables in memory and the pipeline. We need to more intelligently manage the pipeline memory.