Occasional buggy spectrum if array of equal errors given

[leigh2]

I found that if TLS is provided an array of uniform errors then it will sometimes/often cause issues with the power spectrum, and hence (automatic) period selection.
I ran a few hundred tests of white noise models with transits and found that in most cases the short period end of the spectrum had erroneous massive peaks and troughs. Peaks naturally caused an incorrect selection of the period.
I've managed to reproduce this from your quick start tutorial notebook and pushed it to my fork here
https://github.com/leigh2/tls/blob/error_array_bug_example/tutorials/01%20Quick%20start%20with%20synthetic%20data.ipynb
(Note that seeds 0 and 2 are fine, indicating that it only occasionally breaks!)

I've included examples of a couple of other power spectrum plots below at different levels of broken-ness.

This was using the current main branch version, on linux, python 3.11.5 fwiw.