libpetab-python v0.9.0

This release will require Python>=3.12 and numpy>=2.1.

Fixes

• Fixed the mean of residuals and added an option to choose between
  normalized and unnormalized errors in plot_goodness_of_fit
  (by @plakrisenko in #473)
• Made scipy a lazy import, so it is no longer required unless
  probability distributions are actually used
  (by @dweindl in #486)
• Fixed PetabStrPrinter for non-integer rational exponents
  (by @wshlavacek in #489)
• Fixed spurious nan-related warnings in Subplot.from_df
  (by @dweindl in #496)
• Fixed a missing return in the noise distribution merge of
  petab1to2, which caused noiseDistribution to always be NaN after
  conversion
  (by @dweindl in #502)
• Fixed the PEtab v2 extension config model
  (by @dweindl in #506)
• Fixed SBML math parsing by using sbmlmath instead of
  sympy.sympify
  (by @dweindl in #509)

Features

• Added support for PEtab SciML problems, including a dedicated linter
  (by @BSnelling, @m-philipps and @dilpath in #482,
  #505, #499, #498, #511, #513, and #516)
• Added support for BNGL models (by @wshlavacek in #508)
• Added linting of the mapping table, and allowed pure annotation-only
  mapping entries (by @m-philipps and @BSnelling in #480 and #500)
• Minor improvements to the v1-to-v2 converter: warn when dropping
  parameterScale/initializationPriorType/initializationPriorParameters,
  and use the experiments table column order suggested by the PEtab format
  (by @m-philipps in #479)