Generalize OneParticleOperator to NParticleOperator and separate operators from densities

[frieschneider]

The current OneParticleOperator class mixes up operators and densities:
the AO↔MO transformation implemented on the operator level is actually for the
density transformation (although it is used correctly).

For example, in Hartree–Fock (HF):

• the Fock operator transforms as
  F_MO = C† F_AO C = ε

• the HF density transforms as
  D_MO_HF = C† S D_AO_HF S C = n

where S is the overlap matrix and n is the matrix of orbital occupancies.

Thus, operators and densities in HF obey different AO↔MO transformation rules.
It therefore makes sense to explicitly distinguish between operators and densities
and to consistently choose the appropriate transformations, also for MP and ADC.

In addition, this separation allows the computation of trace products of arbitrary
operators and densities in both AO and MO representations, yielding basis-independent
results.

What was done

• generalized OneParticleOperator to NParticleOperator
• explicitly differentiated between operators and densities
• explicitly included Hermitian and anti-Hermitian symmetries
• infrastructure for TwoParticleOperators and TwoParticleDensities