Improve ThOr functor testing with different SIMD backends

The main change here is to add a new algorithm for testing compilation and execution of ThOr functors: ThOr::Functors::Tests::{Evaluate,Initialise}<LHCb::Pr::Fitted::Forward::Tracks>.

This requests (functor cache, JIT) compilation of all supported combinations of different SIMD backends etc. For example:

• JIT/cling scalar (should return scalar::int_v, scalar::float_v, scalar::mask_v)
• JIT/cling scalar POD (should return int, float, bool)
• functor cache scalar (should match JIT/cling)
• functor cache scalar POD (should match JIT/cling)
• functor cache SSE (should return sse::int_v, sse::float_v, sse::mask_v)
• functor cache AVX2 ...
• ...

One variant (...::Initialise<T>) stops there, the other (...::Evaluate<T>) also tests that evaluating all of these different versions give compatible results. The latter option only works if there is a source of input data, or if the functors don't take input data (i.e. they are "void"). So far we have a lightweight generator of fake LHCb::Pr::Fitted::Forward::Tracks, so evaluation of those functors is tested.

Itemised changes:

• Use this new algorithm in a new QMTest
  • Remove some old tests that are now more comprehensively tested by the new algorithm
• Improved with_functors and with_functor_maps mixins, notably they now cope with duplicated property names (so you can configure one property in Python, but have this map onto multiple things in C++ such as different SIMD backends etc.)
• Fix a bug in the COMB (re-assemble a combination object from a composite object) functor that was caught by improved testing
• Improve error message from the ThOr functor factory interface.
• Make SIZE functor cope with being told the explicit type of its target container.
• Use ThOr::Defaults::ALL shorthand in a couple more places
• Add pack_dict helper functor to Functors.utils. This is useful when packing a Python dictionary of functors into the property created by the with_functor_maps mixin. This replaces a Python snippet that had already been copied/pasted to a bunch of different places...
• Add an Sel::Utils::as_arithmetic( SIMDWrapper::scalar::mask_v ) overload
• Take advantage of the improvements to with_functors above to remove duplicated (VertexCut/VertexCutSIMD) properties in CombineTracksSIMD
• Improve an error message in CombineTracksSIMD
• Tweak Python-side ThOr functors:
  • Support passing a (c++ type string, header) pair as a template argument to functors with the default parser.
  • Make Python-side ThOr functors know their 'pretty' name (e.g. RUNNUMBER) as well as their 'pretty' expression (e.g. RUNNUMBER(ODIN=....)).
  • Make ThOr functors print their more succinct versions (in some circumstances..)
  • Add a somewhat hacky comparison operator to BoundFunctor so that printing Python configurables with ThOr functor properties works.

Should be tested with LHCb!2661 (merged) and Moore!577 (merged).

cc: @nnolte @mvesteri @apearce