Do you think it's plausible that 8 MB of source code, consisting of modified open-source libraries, represented a theft of critical Goldman Sachs secrets?
Without knowing how much of that 8Mb was proprietary and what the nature of the proprietary code is it's impossible to say. It could be massive over kill, it could be lenient.
Yes.
One 'trade secret' I worked on was about 30 lines of code, in C#. I later turned it in to about 100 bytes of x86...
Some of the trade secrets in quant models can just be the weights.