Implement proper fix for "Broken deserialization of Lists with offsets" (#248)

[chmp]

The current fix for #248 requires skips the elements by deserializing any skipped values. Figure out an alternative that avoids any unnecessary work. Ideas:

• Add skip to the interface of the deserializers and implement a solution by shifting around the offsets
• Implement random access deserialization
  • This will also allow to deserialize enums with holes or lists with null values with data
  • This would also allow to support lists with non-monotically increasing offsets (although such lists could be ill formed)

Potential design of RandomAccessDeserializer

trait RandomAccessDeserializer {
  /// Get the number of items
  fn len(&self) -> usize;
  /// Deserialize the `i`-th item of this deserializer into the given type
  fn get<'de, T: Deserialize<'de>>(&'de self, i: usize) -> Result<T>;
}

struct Deserializer {
  inner: StructDeserializer,
  next: usize,
  len: usize,
}

// impl Deserializer analogous to OuterSequenceDeserializer

Alternative design

Context:

• the deserialize API fully consumes the Deserializer, no option to hand in additional info
• some APIs require iterator style access (e.g., SeqAccess, MapAccess)
  • but the corresponding impl knows the number of elements, e.g., it can always use a non-iterate design
• current impl uses stacked iterators

Extend serde interface to take an index, add wrapper type around a deserializer and index and calls the deserializer with the given index.

[chmp]

Work started in #260

[chmp]

Implemented with #260