Wrong semantics of .env pronouns

[lionel-]

The .env pronouns currently refer to the environment of the outermost quosure:

inner_quo <- local({
  foo <- "foo"
  quo(.env$foo)
})

outer_quo <- local({
  foo <- "bar"
  quo(identity(!! inner_quo))
})

outer_quo
#> <quosure: local>
#> ~identity(~.env$foo)

eval_tidy(outer_quo)
#> [1] "bar"

I think this is incorrect because inner quosures do not own the dynamic environments of outer quosures. Furthermore, this creates inconsistencies due to quosure inlining:

inlined_quo <- local({
  foo <- "bar"
  quo(!! inner_quo)
})

eval_tidy(inlined_quo)
#> [1] "foo"

It should be safe to fix this behaviour as we didn't document this pronoun in dplyr's programming vignette.