Performance: Improve EdgeList

[lvca]

Currently each vertex has a pointer to a Linked List for out and in edges. All the edges are in this linked list ordered LIFO (last edge inserted is the first). This is not optimal for the following operations:

• count all edges -> browse the entire linked list
• count edges of a specific type -> browse the entire linked list
• traverse edges of one edge type only -> browse the entire linked list
• browse edges FIFO (from the oldest)

We'd need to use v1 (from v0, the original format) with the following improvements:

• store edge count (4 bytes - up to 4B edges) in the head of the linked list
• separate the linked list in multiple ones, one per edge type (or bucket)
• save the pointer to the 1st segment to browse FIFO

[lvca]

This branch works, but it slows down the classic use case with one edge type connected to the vertex. While it provides better "OLAP" responses with counts and faster crossing of edges by properties, we need to investigate if this is really worth the price.

IMPORTANT: Once you open a database with the branch edge-list-v1, the database is automatically migrated to the new edge list structure and you can't go back. So for testing, make a copy of the database folder.