TCGA dataset: different accession IDs mapped to same location?

Question

I'm currently working on TCGA miRNA dataset. After constructing a reads matrix, I'm trying to find the isoform sequence. In my data, I have the genomic location (isoform_coords). I found that entries with different accession ID (MIMAT) were mapped to the same location. I looked at the source files to make sure there were no errors during merging. The data frame looks like this:

miRNA_ID        isoform_coords                      miRNA_region
hsa-mir-143     hg38:chr5:149428962-149428983:+     mature,MIMAT0000435
hsa-mir-143     hg38:chr5:149428962-149428983:+     mature,MIMAT0004599

I searched miRBase using the MIMAT ID and found that MIMAT0000435 and MIMAT0004599 are on different arms (5p and 3p) of hsa-mir-143:

hsa-miR-143-3p MIMAT0000435
hsa-miR-143-5p MIMAT0004599

This seems strange since there should be some distance between the 5p and 3p sequences. Is there an explanation for this?

I need to get the sequence for each entry. I can do this using the Ensembl API. Is there a way I could perhaps do some mapping to get the 5p and 3p arms (so far I just did a manual search but I think there should be a more credible way to do this)? I would like to later analyse the expression patterns of the isoforms in cancer tissue so it would be useful to have isoform IDs as specific as possible.