Smodels

* Small fix in pyhf caching code (results should not change)
* Fixes with parallelisation of resummino cross section computer
* Tiny fix to allow '' without subsequent whitespace as comment in text database

Small bug-fixing release:

* Bug fix for printing signal region combination results
* New cluster algorithm (simpler and more stable)
* Made the pyhf backend configurable. WARNING: if backend specified is not found, we now fall back to numpy!

* first database of the SModelS v3 series, able to incorporate simplified models without a Z2 symmetry
* uses new string representation of simplified model topologies (see https://smodels.readthedocs.io/en/3.0.0/SMSDefinitions.html)
* added results from 11 ATLAS and 6 CMS analyses (hfm=HistFactory model, cov=covariance matrix for SR combination):
- results from ATLAS: ATLAS-SUSY-2018-33 (EM), ATLAS-SUSY-2018-16 (EM+hfm), ATLAS-SUSY-2018-13 (EM), ATLAS-SUSY-2018-09 (UL), ATLAS-EXOT-2019-03 (UL), ATLAS-EXOT-2018-48 (UL), ATLAS-EXOT-2018-06 (UL), ATLAS-EXOT-2013-11 (UL)
- results from CMS: CMS-SUS-21-007 (UL), CMS-EXO-20-008 (UL), CMS-EXO-19-012 (UL), CMS-EXO-16-057 (UL), CMS-EXO-12-059 (UL)
- EMs from recasts: ATLAS-SUSY-2019-08 (MA5), ATLAS-SUSY-2018-42 (LLP repo), ATLAS-SUSY-2018-22 (CM2), CMS-EXO-20-004 (MonoXSMS)

Note that the 4 ATLAS-EXOT and 5 CMS-EXO analyses are resonance searches, while ATLAS-SUSY-2018-13 is an RPV SUSY search. These 10 analyses can only be treated with the new graph-based topology description of SModelS v3.

* added smodels-analyses.json
* Database URLs:
- https://smodels.github.io/database/official300 ("official", in short)
- Database URL for Fastlim results: https://smodels.github.io/database/fastlim300 ("fastlim")
- Database URL for superseded results: https://smodels.github.io/database/superseded300 ("superseded")
- Database URL for non-aggregated results: https://smodels.github.io/database/nonaggregated300 ("nonaggregated")
- Database URL for results with full likelihoods: https://smodels.github.io/database/full_llhds300 ("full_llhds")

Note: The above URLs can only be used with SModelS v3.0 or later. The v3.0 databases can be combined using "path=official+fastlim", "path=official+nonaggregated", etc. in the parameters.ini file as explained at https://smodels.readthedocs.io/en/3.0.0/RunningSModelS.html#the-parameters-file

Changes:

* added resummino cross section computer
* fixed bug in computation of error on muhat, for pyhf likelihoods
(affects mostly the numpy backend)
* small change in initialisation of gradient descent method for computation of
combined mu_hat, to increate robustness of method

SModelS version 2.3.2

Changes:

* fixed bug in initialisation of analysis combination
* smodels version now printed in all output formats

Changes:

* fixed bug for reading QNUMBERS from SLHA files
* small fixes in how pythia6 and pythia8 are built
* small fix in truncated Gaussian llhd experimental feature
* small fix in computation of combined upper limits
* combinationmatrices are now forced to be symmetric
* added isCombinableWith method also for CombinedDataSets
* added a recipe for how to use a combinations matrix
* runtime.nCPUs() now returns number of available CPUs, not all CPUs
* xseccomputer now has --tempdir option
* StatsComputer now has CLs method
* changed default prompt width from 1e-8 to 1e-11 GeV in code