Implement .name_spec = "inner"

[DavisVaughan]

Closes #1988
Branched from #1994

To ignore outer names (but use them in error messages), and retain inner names, see #1988 for rationale

Pretty easy implementation for this, and I'm very satisfied that this can function as a .name_spec that works for the fallback case as well. None of the others do, which can be a little frustrating sometimes when designing fully generic APIs.

---

There's only one thing to note. In vec_rbind() we previously disallowed this:

vec_rbind(foo = data.frame(x = 1, row.names = "row"), .names_to = "id", .name_spec = zap())

i.e. when you set .names_to = "id" to consume the foo outer name and promote it to a column name, you can't also zap() the row names, which is supposed to zap both the outer and inner parts of the row names.

I think this is overly restrictive? I think we should just say ".name_spec only receives an outer name when .names_to = NULL, otherwise the outer name is consumed by the .names_to behavior". We don't have to restrict the behavior of .name_spec at all, and it is a little annoying that we do so right now.

You should be able to do the rbind call above. It's requesting:

• Promote the outer foo name to an element of a new id column
• Zap the inner "row" row name

I think this is a totally valid use case.

Similarly we are now going to allow all of these valid cases:

# Zap `foo`, retain inner row names
vec_rbind(foo = data.frame(x = 1, row.names = "row"), .names_to = zap(), .name_spec = "inner")
# Promote `foo` to `id` column, retain inner row names
vec_rbind(foo = data.frame(x = 1, row.names = "row"), .names_to = "id", .name_spec = "inner")
# Let `.name_spec` handle `foo`, then `.name_spec` retains only inner row names
vec_rbind(foo = data.frame(x = 1, row.names = "row"), .names_to = NULL, .name_spec = "inner")