Gene sets in Salmon output differ between experiments

### Context

I am attempting to download the raw (unnormalized) Salmon counts for a set of human RNA-seq samples that span multiple studies (experiments) and compile them into a single read count matrix. Using the API, I have created a dataset with two samples from each of two experiments ([3b4441fe-1522-4db7-951e-1db9fa9c82bc](https://www.refine.bio/dataset/3b4441fe-1522-4db7-951e-1db9fa9c82bc)), with aggregation set to "EXPERIMENT" and quantile_normalize set to False.

### Problem or idea

The problem is that the read count tables for the two experiments differ in their gene sets (annotation), and thus the tables cannot simply be joined. Of course, a trivial workaround is to subset to the intersection of the gene sets. But it is surprising that the gene sets differ, given that both were processed by Salmon with the same genome build, and there is no mention of a different version of gene annotation being used. When I check the details of the processing of these experiments, it does appear that versions of Salmon used for the two experiments are different (0.9.1 and 0.13.1), but I would not expect that to change the gene set.

### Solution or next step

Ideally, all samples would be processed with the same version of Salmon and the same annotation. If a different annotation was used for different experiments, this should be documented in the processing details for that experiment. If the annotation was actually the same for these two experiments, then some investigation into why the outputs have different sets of genes should be done.

Start with the API dataset 3b4441fe-1522-4db7-951e-1db9fa9c82bc and compare the processing details and gene sets for its two experiments. Check whether Salmon 0.9.1 and 0.13.1 used different annotations; done means identifying and documenting the cause, or confirming a reproducible output discrepancy and its scope.

Written by the indexing model from the issue text.