Improve `elvis.config` validation

[bormilan]

Improve the validation of elvis.config with the following:

• Spot if a non-existent ruleset was given
• Spot if a non-existent value was given to dirs
• Spot if a non-existent value was given to filter

The preferred solution is to make it inside elvis_config:validate/1, just like the other validations.

[paulo-ferraz-oliveira]

Spot if a non-existent value was given to dirs

We already detect if a given section didn't match on anything (dir + filter) combo, but not for all combos. This means that if you have src, test in dirs, but only have src it'll not warn (except if the filter matches nothing). The same for src with filter .erl, .hrl where you have no .hrl.

[paulo-ferraz-oliveira]

I'd like to add to your list, if that's Ok:

1. spot that we only have a key at the top: elvis (and that it's in a 2-tuple format) - maybe we can even soft-deprecate if there's no other options inside that are possible
2. spot that inside that key we only have known keys: config being one, but some might also be from elvis so we'd have to support those (and that they're in a 2-tuple format)
3. spot that (I'm not sure we're doing it already) inside ignore we only have modules that actually exist

All of the above are oxbow code -related, since their job is to trim elvis.config to the bare minimum. Occam's razor ftw!

[paulo-ferraz-oliveira]

I've now considered this as closable in #515 given all considerations are implemented (and then some).

[elbrujohalcon]

I agree.