RSR combination fails to be commutative (sometimes)

[teuben]

Our standard benchmark 33551 is commutative if combined with it's observing twin 33552. I.e.

  SLpipeline.sh obsnums=33551,33552
  SLpipeline.sh obsnums=33552,33551

However, recent data, e.g. 104190 and 104191 (from US-18) is not. This came to light as a US-18 combination showed a different waterfall in the first obsnum when observed alone, and observed in the combination.

Obvious thing to try is not use the pipeline but just a simplified rsr_driver call without all the blanking and badlags corrections the pipeline adds.

[teuben]

my benchmark script to show the issue:

 SLpipeline.sh obsnum=104090 bandzoom=0 restart=1
 SLpipeline.sh obsnum=104194 bandzoom=0 restart=1
 SLpipeline.sh obsnums=104090,104194 restart=1   
 SLpipeline.sh obsnums=104194,104090 restart=1

Then compare 2023-S1-US-18/104090/rsr.wf.pdf with 104090_104194/rsr.wf.pdf - and for chassis 2 and 3 the plot for 104090 will look different.

Comparing the 194 without and with the 090 will show chassis 1 to have an obvious difference.

[teuben]

ugh, the problem goes back to us silently assuming that badlags are independent of the obsnum.

likely solution is to determine the final spectrum from a weighted average of individual obsnum spectra