centralized::Storage never evicts, so a node ID spent on a board that is
gone stays spent. Once all of them are, every board the table has not
seen before is refused a node ID through a UAVCAN_TRACE that compiles to
nothing here - silent on the bus and on the console, so a full table
reads as a dead node.
Boards already in the table still get their cached ID and join fine, so
the bus looks healthy. A test station cycling boards through one flight
controller fills the table and stops allocating for good.
PX4_ERR every 5 s while full: the table does not empty itself, so this
is a fault someone has to clear, not a one-time notice.
Jacob Dahl
a3d5fb11bb