Fix issue with underscores in terms for brms models / ran_vals

[sam-crawley]

For brms models, if a term had an underscore in it, the string splitting got confused for ran_vals.

e.g.

data("sleepstudy")
sleepstudy$Days_extra <- sleepstudy$Days
model.brms <- brm(Reaction ~ Days_extra + (Days_extra|Subject), sleepstudy, chains = 2)
tidy(model.brms, effect = "ran_vals")

This would result in "Days" getting added to the levels column, and "extra" in the term column.

This commit fixes this by replacing strsplit with a regex using str_match_all.

[bbolker]

thanks! I'm hoping I (or you) can modify the existing tests so that they check for this issue (i.e. regression tests). May pull this first anyway ...

[sam-crawley]

Yes, I'll add a test shortly.

[bbolker]

also, looks (based on TravisCI) as though this fails the test here?

[sam-crawley]

Should be good now.