Fix trace processing if number of subplots >= 10

[alyst]

Another fix for multi-subplots case.

Test case (single aesthetic bound to trace):

p <- ggplot( data.frame(x = rnorm(1000), y = rnorm(1000),
                   a = sample(1:20, 1000, replace=TRUE),
                   b = factor(sample(1:5, 1000, replace=TRUE))) ) +
  geom_point(aes(x=x, y=y, color=b)) + facet_wrap(~ a )

ggplotly(p)

With multiple aesthetics per trace:

p <- ggplot( data.frame(x = rnorm(1000), y = rnorm(1000),
                   a = sample(1:20, 1000, replace=TRUE),
                   b = factor(sample(1:3, 1000, replace=TRUE)),
                   c = factor(sample(1:3, 1000, replace=TRUE))) ) +
  geom_point(aes(x=x, y=y, color=b, shape=c)) + facet_wrap(~ a )

ggplotly(p)

[alyst]

Previously the split_by name components were excluded from the trace name generation only if they were followed by the "@%&". That concealed the bug in estimating how many scales should be excluded (split_by scales have to be used, not the original data columns).

[cpsievert]

@alyst I'd prefer

isDiscrete <- names(d) %in% paste0(names(discreteScales), "_plotlyDomain") %||% ""

[cpsievert]

Thanks @alyst. I'm closing this, but you're contributions are much appreciated, and have been quite useful for identifying issues. Some of these ideas have been incorporated into #654.