Provide more helpful feedback on `epi_slide` length/nrow mismatches

[brookslogan]

Currently, we get errors like

Error in `Abort()`:
! The slide computations must either (a) output a single element/row
each, or (b) one element/row per appearance of the reference time value in the
local window.

There are two issues:

1. The error shouldn't be "in Abort()"; we should use something like call = caller_env() in the abort call inside Abort (part of Augment Abort & co.; add rlang condition tutorial, helpers #102 but that's a large, stale WIP), or use cli_abort instead.
2. This doesn't give very helpful info about what computation(s) output the wrong number of elements/rows. We could pack the slide computations list and the expected counts in an error object & refer to them in the error message, and/or print the first one with a violation of the existing pattern.

[dshemetov]

I believe this implicitly handled by #519 (and/or adjacent PRs), by forcing epi_slide .f to return a single row. Are we happy with the error messages around that now?

[brookslogan]

The changes you noted seem like they addresses part 2., and cli_abort addressed part 1, for epi_slide. And for epix_slide allowing any number of elements/rows out solved this message being unclear by eliminating it.