improve anonymous type handling

[SoMuchForSubtlety]

This PR introduces multiple improves in handling anonymous types:

coalesce identical anonymous types

old behaviour:

type Anon1 struct {
	Elem []string `xml:"http://example.org/ Elem,omitempty"`
}

type Anon2 struct {
	Elem []string `xml:"http://example.org/ Elem,omitempty"`
}

type ListType struct {
	Elem []string `xml:"http://example.org/ Elem,omitempty"`
}

type Type1 struct {
	ListType ListType `xml:"http://example.org/ ListType,omitempty"`
}

type Type2 struct {
	ListType Anon2 `xml:"http://example.org/ ListType,omitempty"`
}

new behaviour:

type ListType struct {
	Elem []string `xml:"http://example.org/ Elem,omitempty"`
}

type Type1 struct {
	ListType ListType `xml:"http://example.org/ ListType,omitempty"`
}

type Type2 struct {
	ListType ListType `xml:"http://example.org/ ListType,omitempty"`
}

remove dangling anon types

The elements in the xml tree would not be properly updated because children weren't stored as pointers. this left unused types in the final output (see Anon1 in the previous example).

use numbered names instead of Anon

If an anonymous type can't be named after the parent's attribute, then try with an integer suffix instead of naming it Anon. Eg. if ListType is already in use, the name the next anonymous type ListType1 (if the elements are not identical anyways).