Avoid using tempfile() for ID

[krlmlr]

This changes new_id() to use runif() and digest::digest() to generate reproducible IDs. This is important for testing and resolves rstudio/shinytest#174 and #985 (comment).

[cpsievert]

Thanks for the PR @krlmlr! Although it's definitely a corner-case, this change would break cases where one has (probably inadvertently) reset the same seed in-between adding of new layers.

set.seed(1)
p <- plot_ly()
set.seed(1)
p <- add_markers(p, data = mtcars[1:10, ], x = ~wt, y = ~mpg, color = I("black"))
set.seed(1)
add_paths(p, data = mtcars[11:20, ], x = ~wt, y = ~mpg, color = I("black"))

The reasons have to do with the way plotly works internally -- it uses new_id() to attach an ID to each graphical layer and it's currently not smart enough to deal with non-unique IDs.

I'll contact @wch and see if we can't add some special logic in shinytest to provide a better experience plotly+shinytest experience without such a fundamental change to plotly

[wch]

FYI: it's possible to call snapshotPreprocessOutput() to add an output preprocessing function to clean up the raw output value to something that's more deterministic for testing. See:
https://github.com/rstudio/shiny/blob/6ede019/R/snapshot.R#L10-L19

For an example of usage, see: https://github.com/rstudio/DT/blob/80981d3/R/shiny.R#L116-L124

[cpsievert]

Closing in favor of #1379