Create more regression data for #132 + other test strengthening

[pseeth]

We need more regression data for #132 to make sure we don't break these changes in future PRs.

Namely, these comments in the CR should be addressed:

• Can we add an edge case in the tests where all the source files we're putting into the soundscape are silent? Just want to make sure nothing breaks. The scaling factor will end up at 1e10 in that case, but it should be fine when multiplied by all zeros. But might be a good code path to make sure this regularization is working going into the future.
• I need to add a no_audio case for generate_from_jams, but in the long run we should add no_audio cases elsewhere. Shall we open an issue?
• Should we add regression data for the cases where clipping is happening? Or when peak_normalization is on? Would be reassuring to see regression data for those cases going forward.

[justinsalamon]

Also, we should specifically add tests that combine clipping + reverb, to make sure the two operations work well in for the order in which they are applied (in all likelihood we'll start with applying reverb BEFORE peak normalization)