Unordered close-following agents

As requested by Guillaume Sartoretti - the agents should be able to directly follow each other in any order.

Currently they can only follow in agent_id / handle order, because each agent is moved atomically in that order.

• Positions / cell locks are recorded in rail_env.agent_positions
• when an agent moves, its current cell is unlocked, and the new one locked.
• when the next agent moves, it checks the partially updated table before moving (thus "cross-over" conflicts are avoided)
• it means a higher-id agent can step into the cell vacated by a lower-id one, but not vice-versa.

Implementation

• use / copy existing loop over agents to check speed, valid actions, malfunction, etc
• for each agent ready to move cell (ie whose agent.speed_data position_fraction = 1) create an edge in a DiGraph from the existing cell (row,col) to the new one.
• Check the graph for agent conflicts:
  • self loops mean the agent is stationary and this takes priority.
  • cell swaps appear as loops of size 2. Mark both agents as cannot move.
  • agents trying to move into a cell occupied by a stationary agent are blocked.
  • agents close_following another agent are allowed to move, however this means we need to:
  • check that chains of agents which are close_following are not blocked.
  • A chain of close_following agents can be blocked by a stationary agent, or a cell-swap.
  • a chain of close_following agents can be branching so the whole tree is affected by a block at the head.
• Special rule for two agents both trying to move into the same cell:
  • The agent with the lower index wins
  • Implication: an agent can "butt into" a close-following chain if it has a lower index than the one it will block.
  • Implication: at a branch-point in a merging tree of agents, the agent with the lower index goes first.

In the diagram below,

• a black node indicates a moving agent (really the agent is the edge, and the two nodes are its current and next positions)
• a self-loop edge indicates a stationary agent.
• a red node indicates a cell containing a stopped / blocked agent.
• a purple node indicates a cell containing a stopped agent in an attempted swap-over pair.
• a blue node indicates an empty contended cell (two or more agents trying to move into it)
• a magenta node (slightly brighter purple) indicates an occupied contended cell, where the occupying agent is moving out, but two or more agents are trying to move in.