Make transport roles and routing fully PDL-derived

[Hackatro]

Context

Follow-up to #33.

#33 completed the primary producer consolidation:

• BRA/ARG/USA producers now use one shared producer archetype;
• regional producers and originators are derived from the PDL roster;
• producer input bindings include PDL and sidecar dependencies;
• Rotterdam sourcing is origin-aware;
• map placements are roster-derived;
• the day-45 drought regression confirms that Rotterdam admits US supply while
  Brazil and Argentina contribute exactly zero;
• the full test suite passes: 27/27.

The remaining work concerns transport identity and topology assumptions, not
producer behavior. It was intentionally kept out of #33 so that the already
verified behavioral changes would not be mixed with a larger structural
refactor.

Problem

The runtime still assumes several place-named model lists and roles, including
Brazilian, Argentinian, US, and Rotterdam-specific transport identities.

Examples include:

• _peer_list() selecting peers through place-named role comparisons;
• eight place-named land-transport and sea-lane roles;
• generic agent behavior depending on those role names;
• sea-lane materialization relying on a literal transitional set;
• remaining direct accesses to place-named model lists;
• Environment.step() unconditionally accessing sea_lane_arg.

Goal

Make transport and peer resolution depend on the PDL-derived roster and
topology rather than place-named Python roles or fixed model-list attributes.

Adding or removing an origin in the PDL should not require editing Python
routing code.