Improve rating of difficulty

[config.json](https://github.com/exercism/cpp/blob/main/config.json) assigns a "difficulty" to each exercise.
Here they are:
```
1
hello-world
leap
luhn
space-age
two-fer
2
collatz-conjecture
gigasecond
grains
pangram
raindrops
reverse-string
secret-handshake
3
acronym
allergies
armstrong-numbers
atbash-cipher
binary
difference-of-squares
etl
grade-school
hamming
hexadecimal
isogram
matching-brackets
nucleotide-count
pascals-triangle
protein-translation
rna-transcription
scrabble-score
sum-of-multiples
triangle
trinary
4
all-your-base
binary-search
robot-simulator
5
anagram
bob
clock
meetup
nth-prime
phone-number
prime-factors
queen-attack
robot-name
series
sieve
word-count
6
circular-buffer
complex-numbers
7
beer-song
crypto-square
food-chain
roman-numerals
10
binary-search-tree
say
```

IMHO some of those ratings are wildly off, e.g.:
- luhn is definitely more difficult than hello-world, leap, or two-fer
- I've mentored space-age several hundred times. Many students struggle with that exercise because they member functions have to be const-qualified and they have to choose a type that can represent the age in seconds for all tests.
- pangram and isogram are very similar, IMHO they are equally difficult.
(The only thing that makes isogram a little bit tricky is that the description doesn't restrict the characters to ASCII so one has to be careful not to call `std::isalpha()` or `std::tolower()` with a `char`.)
- Are beer-song and food-chain really that complicated? (Frankly, I never understood the appeal of those exercises.)

I'm sure you can easily spot some more discrepancies.

And a more general question: Are the difficulties rated in comparison to each other or do those ratings have some meaning (e.g. 1=for beginners who just wrote their first hello-world, 4=for those who feel comfortable writing classes and functions, 7=complex tasks with multiple requirements)?