Fix bug causing order of parameters to get changed (i.e. alpha[10] comes before alpha[1]).

Include an argument in ggs_traceplot() in order to simplify the traceplot so
that the printing is not so resource consuming. This also passes to ggmcmc().

  - The previous version included a complete move of one part of the code, which
  in fact produces more problems that it solves.

  - So the move is reverted back and the problem with attributes is solved by
  setting them again.

By allowing ggs() to keep variables passed through the par_labels argument, it
is possible now to use those variables in ggs_caterpillar(), because the ci()
function is able to join them all and keeps them in the final processed
object.

  - Provide a function ci() that calculates credible intervals, for general use
  in ggs() objects, beyond ggs_caterpillar().

  - Adapt ggs_caterpillar() to use the newly created function.

  - Increase the data that passes so that several examples can be used
  (including posterior predictive values, posterior expected binary outcomes).

  - Separate them into different objects for the linear and binary examples.

  - From "data(samples)" to "data(linear)".

  - Improve the examples of ggs_separation() and ggs_rocplot().

  - Affects calc_bin(), gl_unq() and roc_calc().

  - The main object from the ggs() call is no more a data.frame, but a tbl_df().

  - All ddply calls transformed into combinations of group_by() calls. Other
  functions (like cbind(), subset() and rbind() have also been implemented using
  the dplyr style).

  - All melt() and dcast() transformed into gather() and spread() (and even some
  as.data.frame.table()).

  - Deleting the possibility of parallelization. It is not implemented in dplyr,
  althoug the C++ backend that does the real work is clearly faster than the old
  plyr.

  - Major revision of ggmcmc() with a bit of an ad-hoc approach to passing
  attributes when there are more parameters than parameters allowed in each
  page.

  - MCMC output from a linear regression model with dummy data
  (data/samples.rda).

  - MCMC output from a logit regression model with dummy data
  (data/samples_binary.rda).

  - MCMC output and other description variables for a hierarchical / multilevel
  model based on radon example at Gelman & Hill.

I believe the goal here is to remove the unused levels of the factor. droplevels(D.sub$Parameter) should be used for this, because the use of factor(as.character(D.sub$Parameter)) was causing the existing ordering of the levels of Parameter to be lost and to be replaced with a purely alphabetical ordering. This was leading to "x[10]" coming before "x[2]" for example. As a result, in parameters were in the wrong order in many of the ggs_ plotting functions if the "family" option was used.

ggs_caterpillar in ggmcmc had a bug when plotting families of parameters and
those families had labels formed by a subset of one of the other labels.

  - Avoid replicated cases of B object

  - For parameters that do not vary, move NaN Rhat value into NA

  - scaling calculation takes care of NAs

  - lower limit in the plot is the minimum Rhat obtained, not 1 as before

  - Reordering was wrong because levels of parameters were incorrectly sorted

  - In cases of more parameters than entries in the par_labels dataset the
  matching was also wrong

maintain parameter ordering for multipage outputs

function continue anyway.

parameters in each of the bins. The function is now inside ggmcmc itself.

Version bump for 0.5.1

in cases of series not changing values the ar() function gave an error.

  - It may generate an approximate number of bins (probably due to rounding
      error).

  - A new function to generate unequal levels when the bins are not equal is
  necessary.

  - The new function must be exported and its arguments passed to globals.R.

Make rug be FALSE by default: it is only a visual improvement that really does not add too much information to the plot, but it takes a lot of resources.