In the last CLDR release, John removed the (prefix|postfix)@noparse variants of Arabic RBNF due to the inherent testing problems. The other languages didn't get their @noparse variants removed, like Hebrew and Maltese. These @noparse variants existed to address the possibility that they may be used in a sentence where the number moves around. Positionally variant numbers should '''not''' be handled at the RBNF level. They can only be handled at the MessageFormat level in order to get the spacing correct in the entire sentence.
We should get rid of the remaining variants.
It looks like the problem appeared because the LDML conversion tool stopped handling @noparse correctly. Rather than fix the tool, the undesirable data should be removed.