I see what you mean but I'm not sure that is the source of the error. I kinda suspect that the example data set you provided works fine. What I think is happening is the labels between the classifier, test and training data are different at some point and so classification ceases.
Your code (albeit I rarely use R) and approach and data look fine. In fact very professional, the data appears scaled (not shown). It looks good.
Perhaps what you refer to is the row name consistency in labelling against the example data set. For example row name, ARHGEF19, this is a gene ID rather than a NCBI code, viz.
efetch -db nucleotide -format gpc -id "LS482351" | xtract -insd gene gene
Elsewhere you use an AL590644.1 as the label, I assume this is the genomic region for PADI4
efetch -db nucleotide -format gpc -id "AL590644" | xtract -insd source organism
AL590644.14 Homo sapiens
(There appears to be version 14 for this subgenomic locus).
So the labelling is a bit mixed and you can feasible convert between them using efetch and scripting a hash/dictionary. However, I don't think its the problem here because its a random forest classification and a label is a label.
It is possible the package you are using is buggy and some of the labels are not getting parsed properly e.g. with a XXX-1 tag or even if a XXX.1 tag. The example data set uses "nice whole" integers. I don't know about R but if regex approaches are used for label parsing things can easily go wrong (its not a good approach). Just a guess however and its easy to check.