Second attempt at C++ implementation of Ebola model

[pratikunterwegs]

This issue requests a second attempt at an implementation of the ebola model. The previous Rcpp implementation was removed in #140, as it did not offer a substantial speed gain.

Possible reasons for no speed gain, and possible targets for improvement:

• prob_discrete_erlang() could be made more efficient;
• use of stats::multinomial() for a single draw could be replaced with a function wrapping sample() or with a custom weighted lottery function; see e.g. this implemenation in {pathomove};
• use of more efficient book-keeping;
• use of more C++ standard library structures instead of Rcpp vectors (?);
• simplifying the implementation to reduce random number draws (perhaps at the expense of more if-else?)

Note: make sure to use Rcpp-provided random number draws, or to include a wrapper to the R random number generator to take advantage of seed management through R.

[pratikunterwegs]

This issue is a candidate for closing: the implementation of stats::rmultinom() is a very efficient way of drawing from a weighted categorical distribution, written in C; it cannot easily be improved by re-writing in Rcpp. This has been added to the design principles vignette.