FieldSanitizer output can fail validateSubset: only [_-] is stripped

The issue is in the FieldSanitizer's stripCharsRegex in registry/conventions/sanitization.jsonld. Compare its current stripping of only [_-] with the validator's pattern ^[a-zA-Z0-9]+$ in validateSubset. Look at the sanitizeNameUnique and sanitizeAnswerCode functions. The fix likely involves expanding the regex to strip non-alphanumeric characters and possibly adding transliteration for umlauts and ß. Test by running the existing sanitizer on the examples in the table and verifying the output against the validator.

FieldSanitizer strips only [_-] (stripCharsRegex in registry/conventions/sanitization.jsonld), but validateSubset requires names and answer codes to match ^[a-zA-Z0-9]+$. So a sanitized name or code can still be rejected by the validator from the same registry.

v0.1.2 (sanitization.jsonld unchanged on main):

How it showed up: formulaid sanitizes generated questionnaires with FieldSanitizer before validating them (formulaid#17). In a real run, a German model produced aktivitätshäufigkeit and täglich. Both passed through the sanitizer, failed the validator, and sent the run into repair rounds that didn't fix them. formulaid now transliterates to ASCII before calling the sanitizer (formulaid 0ba209e), but the converters use the same sanitizer and have the same gap.

Expected: the sanitizer's output always satisfies the validator's pattern. For example, transliterate umlauts and ß (ä→ae, ß→ss), then strip everything outside [a-zA-Z0-9], not just [_-].