output_results_excel writes a corrupted excel file

[g-hyo]

Just ran into this with GOS.

x <- test_dir(full_path, stop_on_failure = FALSE, stop_on_warning = FALSE)

res_path <- paste0(full_path, "/results_pop_final_clean.xlsx")
output_results_excel(x, res_path)

One of the base checks had a reference to E386 which is no longer in the data. This resulted in the base check spitting out an error:

`Problem with `mutate()` input `..1`. �[31mx�[39m object 'E386' not found �[34mi�[39m Input `..1` is `!E386 %in% c(NA, "", 9)`.`

test_dir didn't actually stop due to this error, and the output produced is a corrupted excel file with a blank sheet like so

Obviously there should be no errors in the first place when running unit tests, but it could be useful to still be able to access the output somehow to quickly diagnose the issue. In a CSV, the error is output.

We could also just have it such that output_results_excel spits an error and doesn't run if there are any errors in the tests (probably the simplest solution)

[kinto-b]

Looks like a character encoding issue. The tidyverse uses UTF-8 characters and applies colouring to their error messages. It would appear that this doesn't play nicely with whatever encoding openxlsx uses.

We can resolve simply by removing non-printable ascii characters from the description. @gorcha How does that sound to you?