This adds a method to ensure that a value is from a set of valid values
and raises an error message accordingly.

- Define Enum types for each of the relevant cases
- Adapt call places that were doing such things manually

- adapt all call sites to use the new API

This mimics the dict.keys() method but returns a list instead of a
dict_keys, for cython performance reasons

- Adding 'get_sequence' on MappingNode to access sequences in a mapping
- Adding 'sequence_at' on SequenceNode to access sequences in a sequence
- Adding 'mapping_at' on SequenceNode to access mappings in a sequence

Using "*_at" in sequences allows us to quickly understand if we are
dealing with a sequence or a mapping.

- 'as_bool()' casts a ScalarNode into a boolean, understanding both
  'True' and 'False' as truthy-falsy values, as per node_get(type=bool)
  behavior
- 'is_none()' allwos checking whether the scalar node contains a 'None'
  value. Since 'None' cannot be used when working with booleans, we need
  to have a way of checking for 'None' when we actually need the
  information of whether the value is unset.
- Adapt all call places to use the new API

- 'get_scalar()' allows retrieving a scalar node from a mapping.
- 'as_str()' casts a ScalarNode into a string (thus removing the node
  information).

Both together, those replace 'node_get(mapping, key, type=str)' but
also allow retrieving the 'Node' itself, which will allow in the future
lazier provenance computation.

This allows to get a mapping node from another 'MappingNode',
replacing 'node_get(my_mapping, key, type=dict)'

Also changes all places where 'node_get' was called like that by
the new API.

PyLint helpfully points out that _GitMirror.fetch() does not match
SourceFetcher.fetch() - it is missing the *kwargs argument. The same for
_GitSourceBase.track().

Move towards being able to enable the 'arguments-differ' error faster
by side-stepping the class design debate and ignoring the error for now.

Change _GitSourceBase.set_ref() argument names to match
Source.set_ref().

Calls to `isinstance` can be particularily costly. Using type() is
much faster. The only known case where the `isinstance` was useful
is for dictionnaries where we would ask for a 'Mapping' instead.

Disallowing 'Mapping' for expected_type considerably speeds up the
calls to this functions.

Also add into NEWS