load_predefined_recognizers(countries=[...]) can't load country recognizers that ship enabled: false, so countries=["in"] loads none #2467

Describe the bug

RecognizerRegistry.load_predefined_recognizers(countries=[...]) can only narrow the recognizers that conf/default_recognizers.yaml already enables. Country-specific recognizers ship with enabled: false, so naming a country does not load them. For 13 countries every recognizer is disabled (au, ca, de, fi, in, kr, ng, ph, se, sg, th, tr, za), and for those countries=[...] loads no country recognizer at all, with no warning.

The docs example says otherwise. docs/analyzer/filtering_by_country.md says countries=["us", "uk"] gives "US- and UK-specific recognizers (US SSN, UK NHS, UK NINO, …)", but UkNinoRecognizer is enabled: false and is not loaded.

To Reproduce

presidio-analyzer from main at 5f5a42e (2.2.366), Python 3.12:

from presidio_analyzer import RecognizerRegistry


def country_recognizers(countries):
    registry = RecognizerRegistry()
    registry.load_predefined_recognizers(languages=["en"], countries=countries)
    return sorted(
        (r.country_code(), r.name) for r in registry.recognizers if r.country_code() is not None
    )


for countries in (["in"], ["us", "uk"], ["au"]):
    found = country_recognizers(countries)
    print(f"countries={countries!r}: {len(found)} country-specific recognizer(s)")
    for code, name in found:
        print(f"    {code}  {name}")

countries=['in']: 0 country-specific recognizer(s)
countries=['us', 'uk']: 7 country-specific recognizer(s)
    uk  NhsRecognizer
    us  MedicalLicenseRecognizer
    us  UsBankRecognizer
    us  UsItinRecognizer
    us  UsLicenseRecognizer
    us  UsPassportRecognizer
    us  UsSsnRecognizer
countries=['au']: 0 country-specific recognizer(s)

Expected behavior

In #2000, which added the filter, the agreed rule was "calling with a country code returns all recognizers with this country value + all recognizers with None". I expected countries=["in"] to load the six India recognizers (InAadhaarRecognizer, InPanRecognizer, …) and countries=["us", "uk"] to include UkNinoRecognizer, as in the docs example.

Additional context

RecognizerListLoader.get skips disabled entries while building the instances (recognizers_loader_utils.py:412 and :439) and applies the country filter to what is left (:466). A country that is named can't bring back a recognizer that was dropped earlier.

The filter was probably more useful when it was added. Since then, CONTRIBUTING has required new country-specific recognizers to ship enabled: false, so each new country makes the gap bigger.

I see two ways to fix it, and I'd like to know which you prefer before I open a PR:

1. Naming a country enables it. When countries is passed to load_predefined_recognizers, load the named countries' predefined recognizers even if they ship enabled: false. This matches the docs and the rule from feat(analyzer): add optional country filter to load_predefined_recogn… #2000. It changes behavior: anyone already passing countries=["us"] would also get the 9 US recognizers that are disabled today (UsMbiRecognizer, UsNpiRecognizer, AbaRoutingRecognizer, …), which can add false positives. A user's own YAML with an explicit enabled: false should keep winning. Only the shipped defaults would be overridden.
2. The filter stays narrow-only. Fix the docs example and the docstring to say the filter only narrows what is enabled. Log a warning when a requested country ends up with no recognizers, and point to enabling them in a registry YAML. No change in detection.

I found this while checking which of the existing India recognizers I could use instead of the custom Aadhaar and PAN recognizers in a project that redacts Indian records. I'm happy to send the PR once you pick a direction.