Add HPO support in SygnalInterfaceSocketcan via parse-ordering arbitration

[eric-polymath]

Summary
Wires SygnalHpoInterface (landed in the prior PR) into SygnalInterfaceSocketcan and the ROS 2 node so HPO boards can be commanded and observed end-to-end. Tactical fix for now, proper per-board refactor + multi-protocol PRD deferred.

Detailed change log

Lib:
New HpoId struct + SendHpoCommandResult; constructor extended with optional hpo_ids (defaults to {}, zero behavior change for existing callers).
sendHpoControlEnable / sendHpoControlCommand + get_hpo_interface_states / get_hpo_overall_interface_state accessors.
parse() rewired: HPO parsers run before the existing MCM control-response parser, with a DO NOT REORDER comment block explaining why.
Constructor throws std::invalid_argument if any HPO and MCM share a bus address.
HPO_NUM_INTERFACES corrected from 7 → 5 to match real hardware (DBC defines 7, only 5 are wired). Matches MCM precedent.
ROS 2:

New hpo_endpoints:

int_array parameter (default []).
parse_hpo_id helper + hpo_ids_ member; passes vector through to the socketcan ctor.

Tests:
New always-on sygnal_interface_socketcan_routing_tests (6 cases / 29 assertions): disjoint-address ctor check, HPO/MCM heartbeat routing, HPO control-response non-disturbance of MCM queues.
Existing HPO interface tests updated for the 5-interface array length (17 cases / 74 assertions, green).
Why this works (CAN-ID arbitration mitigation)
Sygnal overloads CAN IDs across device types — 0x161 ControlCommandResponse carries different byte layouts depending on whether MCM or HPO produced it. The only field with stable semantics across layouts is the 7-bit BusAddress. We exploit two invariants:

Routing order :

parse() tries HPO parsers (which filter by bus_address) before SygnalControlInterface::parseCommandResponseFrame (which doesn't filter). Anything addressed to an HPO is claimed first; what reaches the MCM parser by elimination is MCM.
Disjoint bus addresses — enforced at construction time so the routing-by-elimination is sound.
What's deferred
Proper per-board refactor (move MCM control parse/pack into SygnalMcmInterface, delete SygnalControlInterface).
IO board, relay refactor, error-frame surfacing.
Multi-protocol/multi-version PRD (Emerson's thread).
Count8 rolling counter — still hard-coded to 0.
Breaking changes
None — existing two-arg constructor still works; mcm_endpoints param unchanged; legacy MCM behavior untouched.

Test plan
colcon test --packages-select sygnal_can_interface_lib green
colcon build --packages-up-to sygnal_can_interface_ros2 clean
On LEV05: send HPO ControlEnable + ControlCommand, observe heartbeat reflecting the command
Confirm misconfigured hpo_endpoints overlapping an MCM throws at node configure