- Remove #nullable disable and add a couple of annotations.

- Add IsEmpty property to MemoryBuilder<T>
- Add Push(), Peek(), Pop, and TryPop() methods to MemoryBuilder<T>
- Change AsMemory() to return a Memory<T> rather than a ReadOnlyMemory<T>.
- Mark indexer as read only.
- Add unit tests.

Add GreenNode.GetEnumerator() method that returns a ref struct-based Enumerator to perform a depth-first traversal using the target GreenNode as the root. SyntaxList nodes are transparent and not returned from the enumerator.

Unit tests are included along with some helpers to make constructing GreenNodes a bit easier. The helpers include:

- A handful of SyntaxFactory methods to avoid passing extra arguments.
- A [CollectionBuilder] attribute for SyntaxList<TNode>
- SyntaxList.Create(...) methods for the [CollectionBuilder] to target.

Add GreenNode.Tokens() method that returns a TokenEnumerable, a ref struct that provides a GetEnumeartor() that returns a ref struct-based Enumerator to perform a depth-first traversal that returns SyntaxToken instances. Unit tests for Tokens() and TokenEnumerable.GetEnumerator() have been added.

This change removes the StringWriter machinery from GreenNode. It only adds overhead. In its place is a more efficient implementation using string.Create. This avoids copying the characters into a StringWriter and then allocating a new string from the writer, which copies all of the characters *again* into a new string.

Because we know the width of a GreenNode, we know the length of the string it will produce.

string.Build(...) collects "chunks" of text as ReadOnlyMemory<char>s before creating a string from them. This ensures that characters are not copied to a buffer and copied again when allocating the final string.