Conversion between eras with NA units

`yr_transform()` currently throws an error if either the source or destination era has an NA unit, following the R convention that `NA == NA` is NA.

It might not be necessary to be this strict, for example to allow transformations between era("bp") and era("bce"), and just warn that there is an assumption that the units are comparable.

Start with yr_transform() and inspect its current handling of source and destination eras with NA units. Define the expected conversion and warning behavior for cases such as era("bp") and era("bce"); the work is done when those cases and existing conversions have clear tests.

Written by the indexing model from the issue text.