overhaul format conversion functions

[corybrunson]

The convenience functions is_*_form() and to_*_form() need a few upgrades:

• They don't currently accept tidyselect specifications.
• They still use the weight parameter, whereas the layers have switched to y, which may cause confusion.
• Internally, to_*_form() rely on tidyr::gather() and tidyr::spread(), which are stable but outdated.
• to_lodes_form() cannot handle alluvial data with variable y (and weight) values. The new function tidyr::pivot_longer() makes this possible, as illustrated in Longitudianal data  #51. It should be possible to similarly convert from lodes to alluvia form while preserving y (and weight) values using tidyr::pivot_wider().

For consistency between these functions and the layers, it may be prudent to deprecate axis[0-9]+ in favor of x[0-9]+, which would better harmonize with the natural y[0-9]+ that would be used to specify variable y values.

[corybrunson]

Because the key and value parameters came from tidyr::gather() and tidyr::spread(), whereas some, probably all, of the internal pivoting will be switched to tidyr::pivot_*(), this would be a good time to deprecate these and the id parameters in favor of the x, stratum, and alluvium parameters used by the layers.

[corybrunson]

There is now a whole book on tidy evaluation, which may be useful for improving the readability and efficiency of the parameter change.

[corybrunson]

Bump.

Also, as the established to_*_form() functions may take a while to deprecate, their tidyr::pivot_*()-based superseders might better be given new names, e.g. pivot_to_*(), so that both could exist simultaneously.

[olivroy]

I understand. I could take a look. There are some useful docs on the rlang website. I think it is worth removing the dep on lazyeval to make things more consistent.

I also spotted a defunct function: dplyr::vars_select() in the ggalluvial codebase. its stated replacement is tidyselect::eval_select().