feat(errors): improved error handling via error classes, encpapsulation returns the condition objects

[sebffischer]

This pull request improves errors and warnings by:

1. Introducing proper error classes:

2. Improving the formatting of error messages to include the error class. Because we now let the user react differently to different error classes (for now only during encapsulation), it's good to make the user familiar with the error class system. For formatting of errors, we use cli, which was already a hard dependency before.

Below are some examples for how error messages and warnings are now formatted:

mlr3misc::error_timeout()
#> Error: 
#> ✖ reached elapsed time limit
#> → Class: Mlr3ErrorTimeout
mlr3misc::error_config("Passed in regression task to classification learner")
#> Error: 
#> ✖ Passed in regression task to classification learner
#> → Class: Mlr3ErrorConfig
mlr3misc::warning_config("this was maybe wrong")
#> Warning: 
#> ✖ this was maybe wrong
#> → Class: Mlr3WarningConfig

To create the condition object, set signal = FALSE:

mlr3misc::error_timeout(signal = FALSE)
#> <Mlr3ErrorTimeout: 
#> ✖ reached elapsed time limit
#> → Class: Mlr3ErrorTimeout
#> >

TODOs:

• what do we want to do with stopf(), warningf(), messagef()?
  --> they keep the old behavior for now. Maybe we slowly want to phase them out.
• class starts with M
• rename silent to signal
• don't wrap messages in condition objects when returning them from encapsulation

Note that encapsulate() currently returns duplicate information, but this is only for the transition phase, see #142

[sebffischer]

Should we allow to include the call?

not include it for now.

[mllg]

Maybe move the exported helper functions from mlr3 to misc to not polute the namespace