Template file added to Sconscript to autogenerate code in
solution.py.in

These examples use the units module to implement the existing examples,
but demonstrate how alternative units can be used for the input and
output quantities.

Since the only overridden method was __init__, it makes more sense for
these to be functions than subclasses.

* To allow setting properties on the upstream classes, __setattr__ is
  implemented and checks whether the attribute is defined on this
  class. This requires setting the phase instance directly into the
  __dict__ attribute to avoid a recursion error.
* Class inheritance is not used here because it is not easy to set
  attributes on the super class. By design, super() does not allow
  setting properties, only getting. Furthermore, since the base
  PureFluid class is defined in Cython, the attributes of that class
  cannot be set by subclasses at all, again by design.
* The PureFluid class implemented here does not inherit from the
  Solution class implemented here because eventually this Solution class
  will include methods and properties related to kinetics and transport,
  which are not implemented for PureFluid. The base PureFluid class is a
  subclass of ThermoPhase only, whereas the base Solution class is a
  subclass of ThermoPhase, Kinetics, and Transport.
* To reflect the distinction between Solution and ThermoPhase, the
  template variables are renamed.
* Several methods of the PureFluid are getters only, despite having
  three properties. These attributes are fixed here. TPQ is the only
  three-property attribute with a setter.
* Setter methods now raise a CanteraError if the unit conversion to base
  units fails due to an AttributeError.

The existing assertArrayNear method was causing units to be stripped
from the array Quantities. Avoiding numpy.asarray conversions and
using numpy.isclose prevents that from happening.

* Remove unnecessary tests of the underlying implementation that
  duplicate functionality in the upstream tests.
* Move some dimensionality tests into the PureFluid tests where the
  saturation and critical attributes exist.
* Remove unused imports

This prevents KeyError failures if sphinx_docs=y and python_package=n

This supports pickling and unpickling of quantities carrying Pint units.

The include_package_data config option is redundant with explicitly
listing the package data, so setuptools was generating warnings. Since
we explicitly list everything, this can be removed.

The units package shadowed the units Cython module. This made Cython
very unhappy and we couldn't import anything from Cantera.

This lets us check whether the wheel is built properly and includes all
the files necessary for the package to function.

Allow setting the backed of the Water instance. Fix a typo in the R134a
instance.

Rather than checking each value individually with try/except, use the
ito method and try to convert within a loop. This simplifies testing for
the error condition and looks a little cleaner.

This increases the line coverage to near 100%, uses pint's testing
assertions rather than rolling our own, and uses parameterize
extensively to reduce boilerplate in testing as many properties as
possible.

Most of the docs are copied from the upstream object. Some updates are
included for units.

This is required to find the version of Cantera in the dist folder. By
default, pip only finds stable versions, and in this case, was going out
to PyPI to find Cantera rather than use the local artifact. However, we
can't set --no-index because we also need to install other dependencies
from PyPI.

Ensure that we only install the wheel from the archive and don't
accidentally reach out to PyPI.

Co-authored-by: Ray Speth <yarmond@gmail.com>