Test files include 0, 1 and 2 mwfn types. The original fchk
were taken from iodata/test/data and converted with Multiwfn3.7

1. The check for Wfntype, to make sure that wavefunction type is of
   legitimate type, is moved to _load_helper_opener function, because
   the error message raised (in case Wfntype is not known), also shows
   the line number where parsing failed, so it is more helpful to add
   the check right after Wfntype is parsed.
2. The 'mo_kind' value is added to the data dictionary, to avoid the
   need for checking Wfntype again.

1. Checking data['Ncenter'] & data['atnums'] are moved to the functions
   which load them, because it is more helpful to raise the error where
   the information is read, as the error message also shows the line
   number causing error.
2. Checking data['shell_types'], data['shell_contraction_degrees'], and
   data['shell_centers'] length are removed, because that would be true
   by construction.

This allows for better checking the title of the section being parsed,
and makign sure primitive exponents & coefficients are correctly assigned.

Unlike other formats (like WFN), itt is pretty simple to build
MolecularBasis from parsed MWFN data, so the _build_obasis func is
removed. This makes the code more clear and easier to read.

1) The raised message would be more informative that the assert error
2) The `assert len(section) == nprim` is redundant, because the while
   loop gaurantees that it would have the same number of elements as
   nprim.

Most of the items in the extra dictionary were already existing amon
IOData attributes, so they were removed.