(Experimental) PID sync, axle-based remote steering, cluster sync for coupled vehicles and other things

[spectrachrome]

Trailer & coupled-rig multiplayer sync

Added

• Full coupled-rig synchronization for trucks, trailers, and chains.
  Coupled trailers run on their physics coupler constraint while the
  GE side computes an expected hitch-point position for drift detection
  and debug visualization.
• Hitch-type branching for sync behavior: fifth-wheel (yaw-only
  articulation), ball hitch (3-DOF quaternion), and pintle (loose
  3-DOF with wider drift tolerance). Classified once at attach time
  from the coupler tag and cached on the coupling.
• Three-way coupler event capture to close BeamNG event-dispatch gaps:
  per-vehicle kiss_couplers.onCouplerAttached hook, GE-level
  onCouplerAttached dispatch, and onObjectCouplingChange diff of
  attachedCouplers as a belt-and-suspenders fallback.
• Time-based dedup on outbound coupler events so the three capture
  paths don't send duplicates.
• Inverse-index tracking (coupled_trucks) so GE-side sync can look
  up a truck's trailers and know when to skip try_rude or apply
  coupled-specific scaling.
• Cluster-aware teleport: when any member of a coupled cluster drifts
  past the threshold, the whole cluster is translated rigidly to
  preserve relative pose and keep the coupler constraint intact.
• Graph walk over coupled_to / coupled_trucks so the cluster
  discovery works for road trains transparently.
• Per-vehicle mass reporting from vehicle Lua to GE so coupler sync
  can size PD gains by the truck/trailer mass ratio at attach time,
  plus a mass-based scaling of the coupled rotation gain itself.
• 3-second decouple grace period after coupled_to is cleared,
  skipping full PD sync during the brief detach→re-attach window
  (e.g. after a reset) so the physical coupler isn't ripped before
  the re-attach event arrives.
• kiss_electrics.set_coupled state tracking on vehicle side so
  advanced coupler electrics only fire when actually coupled via
  our tracking. Remote advanced couplers are force-detached at
  spawn to prevent auto-attachment to nearby objects.
• Retry loop for tryAttachGroupImpulse over a 500 ms budget so
  advanced couplers reliably retract landing gear after the async
  C++ latch completes.
• Parking brake release on the trailer at attach time, since trailers
  don't send VehicleUpdate and wouldn't otherwise get input sync.
• coupler_tag field on CouplerAttached network message so the
  remote can classify hitch type from the same event.
• Axis-swizzle conversion in the coupled-truck PD so network angular
  velocity (packed pitch, roll, yaw) is subtracted against local
  angular velocity (built yaw, pitch, roll) in matching slots.
• Debug visualization on the vehicle-side PD (force vectors, drift
  line, angular-force indicator, expected/received/current spheres,
  drift threshold markers) gated by M.debug.
• Tick-paced debug logging of per-vehicle sync state gated by
  M.debug_log.

Changed

• Coupled-truck rotation PD now uses coupled_ang_force = 2 (1/50 of
  the non-coupled gain) with a hard cap of 3, giving a proper
  restoring force on all three axes without whipping the hitch
  per tick.
• try_rude (6 m per-vehicle teleport) is skipped for coupled trucks
  because it would yank the rig apart under trailer drag lag. Drift
  is handled cluster-wide instead.
• kiss_gearbox.apply is always called on synced vehicles; the former
  if not coupled_to[id] gate was removing gear sync from the truck
  side after coupling.
• Ownership guard on inbound coupler attach/detach events now checks
  both vehicles, not just obj_a, so rebroadcast events don't
  reposition our own rig.
• Inbound attach_coupler is idempotent on re-received events for the
  same pair so duplicates don't disturb the active constraint.
• Vehicle spawn cleanup: known_couplings, coupled_to, and
  coupled_trucks state is cleared on mission load and vehicle
  destruction.

Fixed

• Jackknife on coupled rigs caused by the full non-coupled rotation
  gain (100) whipping the hitch node per tick.
• Second trailer in a chain flipping around its longitudinal axis on
  asymmetric drives — network yaw rate leaking into the remote's roll
  slot via mismatched angular-velocity packing.
• Coupler ripped out of the truck on every reset, often throwing the
  chain too far to re-latch.
• Truck gearbox permanently frozen on the remote at whichever gear
  was active at attach time.
• Articulated bus and other long rigs breaking themselves apart /
  flipping at speed, from an experimental hitch-pivot rotation that
  injected energy through the wrong axis.
• Light rigs over-rotating and heavy rigs under-rotating because the
  coupled-truck rotation gain was mass-invariant.
• Owned vehicles yanked sideways by rebroadcast coupler events bouncing
  back to the sender.
• Angular acceleration never clamped (the guard was checking linear
  acceleration twice), causing big ang-accel spikes to pass through
  unfiltered on lossy networks.
• Self-coupling events from fifth-wheel hitches without a trailer
  being sent as real couplings.
• Ball-hitch L-key couplings that the vehicle-side callback missed
  now land via the GE-level onCouplerAttached dispatch or the
  onObjectCouplingChange diff.
• Advanced coupler electrics grabbing random nearby objects on
  remote vehicles at spawn, since their notAttached state was
  being driven by synced values.

Known issues

• Systematic trailer drift under sustained correction.
• Cars and loose cargo on tilt decks sliding off, mostly downstream
  of trailer drift.
• Articulated bus back-end flex and doors popping under PD forces.
• Low fifth-wheel hitches detaching on reset because the trailer's
  landing gear drops below the hitch height before the coupler
  can re-latch.

[florinm03]

Cargo.lock and claude is not needed on the repo, make sure to only push what's necessary ;)

[spectrachrome]

Sorry, of course! I'll ditch the CLAUDE.md and lock file in the next commits and generally clean this up. Old habits :)

Also going to look into the many merge conflicts.