[PoC] Can we really use tibble internally?

[yutannihilation]

❗️ Do NOT MERGE THIS ❗️

This is an attempt to use tibble internally as discussed on #3018, but currently I feel this is going to be a tough job. Although this is still at the very early stage of work in progress, I already see several dificulties:

• $.tbl_df warns when there is no such column. So, everytime we extract a column by $, we need to check if the name exists rather than check the result is NULL (c.f. 'unknown or uninitialised column' warning is annoying tibble#450).
• Not all functions respect the tbl_df class. For example, transform() and cbind() returns data.frame. Besides, in some places the users can supply arbitrary functions, which meams we have to convert the result from those user-input functions to tibble very frequently. I'm afraid this not only does harm on performance, but also makes the code error-prone.
• Many unexpected test failures...

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/