Update to incorporate new features in drake

[wlandau]

Development drake now tracks the history and provenance of targets over time (CRAN ETA mid-July) and the additions in this PR describe how it works.

drake history is similar to MLflow tracking except:

1. You do not have to tell drake which parameters to track. drake_history() just analyzes commands automatically and detects named length-one atomic arguments to function calls in the plan.
2. drake does not automatically create downstream summaries such as performance metrics. Users still need to create special targets for those.

[wlandau]

As of ropensci/drake#977, we no longer need to worry about the performance issue I mentioned in richfitz/storr#77 (comment). With target(format = "keras"), drake can now save and load Keras models directly in HDF5 format instead of a storr RDS file. The workaround from #9 is obsolete, so I removed it in b982bf9 when I added target(format = "keras") just now.

Related: drake also supports fst storage for large data frames. It does not immediately work for the data in this example because it is an object of class rsplit, but a little refactoring might help us work around this little technicality.

library(drake)
n <- 1e8 # Each target is 1.6 GB in memory.
plan <- drake_plan(
  data_fst = target(
    data.frame(x = runif(n), y = runif(n)),
    format = "fst"
  ),
  data_old = data.frame(x = runif(n), y = runif(n))
)
make(plan)
#> target data_fst
#> target data_old
build_times(type = "build")
#> # A tibble: 2 x 4
#>   target   elapsed              user                 system    
#>   <chr>    <Duration>           <Duration>           <Duration>
#> 1 data_fst 13.93s               37.562s              7.954s    
#> 2 data_old 184s (~3.07 minutes) 177s (~2.95 minutes) 4.157s

^{Created on 2019-08-05 by the reprex package (v0.3.0)}