Feature request: Moroccan recognizers (MA_RIB, MA_CIN, MA_CNSS) #2474

Is your feature request related to a problem? Please describe.
Presidio has no recognizer for Morocco, and no predefined recognizer that declares Arabic (ar). Moroccan text, in Arabic or French, carries identifiers that nothing detects today: the bank account identifier (RIB), the national identity card number (CIN) and the social security number (CNSS).

Describe the solution you'd like
Three country-specific recognizers under predefined_recognizers/country_specific/morocco/, shipped disabled, for languages ar and fr. I have them written and tested, and I would send them as three small PRs, one recognizer each.

Sources:

• RIB structure: U.S. Social Security Administration, POMS GN 02402.297.
  The key makes the 24 digits divisible by 97. No official text states thisrule; it holds for the RIB that Bank Al-Maghrib published for the special fund no. 126, and for three other published RIBs from three other banks. Because about 1 random number in 97 also passes, the key only removes candidates and never raises the score.
• CIN: law no. 04-20 (name of the number), the World Bank ID4D country diagnostic for Morocco (two letters + six digits + two check digits), and the NARSA road-offences portal, which accepts one or two letters followed by one to six digits.
• CNSS: the same World Bank report (nine digits, leading 1, two control digits).

Each recognizer comes with a configuration-path test through RecognizerRegistryProvider, exact score assertions, lookalike negatives, and context tests in ar and fr.

Two things I would like your view on before I open the PRs:

1. These would be the first ar entries, so ar joins the known languages asserted in test_recognizer_registry.py. Is that welcome?
2. No Arabic or French model is installed in CI. For context tests I use a small test helper that tokenizes with spaCy's rule-based tokenizers (spacy.blank). Is that acceptable in the test suite?

Describe alternatives you've considered

• Promoting a RIB to 1.0 when the key passes, as IbanRecognizer does. I chose not to: a bare 24-digit number has no country prefix, so a filterable score seemed safer.
• Using the labels "RIB", "CIN" as default context words. I left them out: context matching is by substring, and they sit inside common French words ("tribunal", "cinq", "médecin").
• One PR with all three. I prefer three, so each can be reviewed on its own evidence.

Additional context
I also looked at the Moroccan passport, the company identifier (ICE) and the driving licence. I am leaving them out for now because I could not find a source strong enough for the format or the check rule.