Not necessarily true. The foregut may simply not be the right environment for them to "stay around", so they just keep moving through the gut until they find a region that they can do well and become enriched there (midgut). Or perhaps it was indeed diet, and you happened to sample at a point and time when the DNA from the Cyanobacteria had moved to the midgut. Of course, these are just a few thoughts 
. Sounds like you may have an interesting research question blooming. 
Remember trying to identify very specific taxa at the genus and species level can be difficult with amplicon sequencing data. It could be there are no good representatives of non-photosynthetic cyanobacteria within the reference database. I've not looked very thoroughly myself. Though you can look here. Note, I did not find any Melainabacteria within SILVA. So they may be present under another name or may simply not have been included yet. Thus, it is possible that the assigned taxonomy is simply a spurious result of the query sequence consistently mapping to the closest, but unrelated, taxon within the reference database. Also, keep in mind that taxonomy is always changing, and new sequence data is being generated.
You can always use the various tools within QIIME 2 and RESCRIPt to fetch and append any 16S rRNA gene data of Melainabacteria from GenBank into your existing SILVA or other reference database (Assuming the taxonomy labelling schema is constant between the files, i.e. GenBank and SILVA differ in some respects. But you can use qiime rescript edit-taxonomy ... to help). Then you can merge the new data into the existing reference database by using the qiime feature-table merge-seqs ... and qiime feature-table merge-taxa ... commands. Then you can train your new classifier.