Parallel computing with parLapply

• Parallel computing with parLapply.
• Improvements to the documentation, including a "caution" vignette.
• The tracked() function, which lists the objects in a workflow that are reproducibly tracked.

Tiny fixes in documentation and a test.

The actual package wasn't the problem, but the docs and testing were:

1. The quickstart vignette called rmarkdown::render(), which created CRAN package check warnings for solaris. I commented out my_render() in the workflow data frame in vignettes/quickstart.Rmd and inst/examples/basic/basic.R.
2. Intermediate file test was intermittently failing because of #4, which is a known issue, but not a show-stopper. Downsizing the size of input test files (toggling a comment in debug.R) fixed the test. All I did was move a hash symbol from one line to the previous one in debug.R.

Infractructure cleanup

• Infrastructure: I reimplemented most of the main functionality with cleaner code, having learned from many lessons and surprises from initial development.
• Robustness: I am using safer mechanisms to search for dependencies and map out user-side workflows so that drake is less likely to fail on edge cases.
• Documentation: I refactored the vignettes, and I baked an example of a common use case into the code.
• High-performance computing: distributed computing with Makefiles requires expensive overhead, so I added a low-overhead single-node alternative (using parallel::mclapply()).