Broken Line Fit for LST

[VourMa]

This PR includes a prototype implementation of the Broken Line Fit for the LST TCs, aiming to produce Tracks directly from the LST producer on GPU. The description of the first working version of the development was presented in these slides:
https://indico.cern.ch/event/1679835/contributions/7060909/attachments/3316119/5935962/BLFForLST_20260717.pdf

The performance is being tested on CMSSW_17, due to availability of samples, and the commits have been cherry-picked (alongside potential, small adaptations) to CMSSW_20 (this PR). I have tried to keep track of the performance for each developments, based on the CMSSW_17 branch commits. The performance corresponding to the slides above is included in in the plots from the first working version. As discussed in the meeting, a fair comparison can be made if I removed the pLSs from the "baseline". This comparison is included in this link and the result is can be found below.

Click me for performance plot!

Since then, the following updates have been done:

• The miscounting of the number of OT hits has been fixed, without any impact on the performance (plots).
  Click me for performance plot!

  
• I realized that there was a redundant selection on the TC type on top of the selection on the number of TC hits. I removed it and split the kernel instantiations to speed up the compilation time. This ended up being a technical change (as I was covering all TC type and number of hit combinations properly), so the performance stayed unchanged (plots).
  Click me for performance plot!

  
• As per suggestion, I included all hits (two per layer, not one, as before) in the fit. That lead to a huge degradation of the performance (plots).
  Click me for performance plot!

  
• As part of the development above, I realized that including two hits per layer requires hit ordering, as this is r-ordering is not guaranteed by LST code. I implemented this but recovered only a small part of the inefficiency that was introduced by including all OT hits in the fit (plots).
  Click me for performance plot!

  

I am making this PR to gather feedback on the code implementation.

Known issues/future developments:

• Including IT hits.