vec_equal_na should use all for data frames

[hadley]

Not any, as currently.

Consider the case of dropping missing values — you only want to drop rows where all values are missing. This will make rcrds and data frames different, which I think makes sense (since a rcrds row represents a single value, but a data frame row represents many)

[hadley]

But we can't make data frames and rcrds behave differently here because the comparison happens via vec_proxy_equal(). But maybe that's ok — if a record wants to behave differently here, it should probably be providing its own proxy.