Problems when recomputing var lss in true continuum analysis

[andreicuceu]

In some mocks the true continuum analysis produces very strange results when recomputing var lss. See attached images of mean continuum and correlation function with and without this. My best guesses are that either it has something to do with the bin at the edge of the lambda grid having fewer pixels, or with var lss being negative in high noise regions (with little data) as #927 pointed out. However, I've tried reproducing it manually in a notebook, and those results looked fine. It's even stranger that it only appears in some of the mocks (2 out of the 5 I've ran).

As it's not clear why this is even needed, the option has been turned off by default in #1008, and a warning was added when using it.