If you publish your software, you are required to release everything needed to develop it further, yes.
However, if you publish your paper, you are not required, under the free software principles.
If I am wrong, show me where the FSF/GNU say that.
I believe your argument is that there's a clear parallel to publishing a paper and publishing the software.
However, there's also a clear parallel to, say, publishing an internet service based on free software, and publishing the software. This lead to the AGPL.
The GPL does not have that requirement, which means it isn't a free software principle. It could have - I think I've heard Eben Moglen say it was discussed early on. But if so, then having a telnet server, or ftp server, or anything reacting to the network, would require the ability to download the source.
> free software people have been working out these "reproducibility" issues in practice for decades before the biologists even got on the bus.
While biologists were working out other "reproducibility" issues long before there were buses.
My argument is that that analogy doesn't follow from free software principles.
If it did, then it would follow that if you publish the paper, you're obligated to release your source.
And we don't see that coming out of the free software movement. We do see that coming out of the open data/open science movement.