ID field.
It's really not a sustainable approach to try to address nodes on a CAN bus by serial number, though. CAN is content addressed rather than receiver addressed. Due to the way arbitration works on the bus, it's invalid for two nodes to transmit to the same CAN identifier. The arbitration mechanism breaks down and results in error frames, at which point the CAN bus is in a degraded state.
That would preclude a CAN enabled bulb from being able to send telemetry back, at least until the bulb was provisioned an identifier. That could be done by an ECU sending a frame with the bulb's serial number and assigned identifier. You still need a zero-conf discovery protocol, though, and so you're back to transmitting before provisioning. You could work around all that, but it's a lot of work.
Stepping back a bit, running a car's CAN bus over a light bulb socket is going to cause some practical reliability problems. Compared to a wire harness going into an ECU, a user serviceable bulb socket is going to be much more prone to intermittent connections from vibration, as well as oxidation and wear. Intermittent connections on CAN_H/CAN_L tend to cause a ton of frame errors, and significantly degrade the overall bus performance often to the point of system failure. When a node encounters enough error frames, it is compelled by the standard to go into a BUS-OFF state where it isolates itself from the bus. Because it's a bus and all the nodes share the same two wires, it's pretty much impossible to diagnose where an intermittent connection is without trial and error.
Ie, with CAN, each node only needs to do reg reads/writes/datasheet-spelunking for a narrow part; the other nodes just need to know the API that sits on top of the hardware.
See for example https://github.com/commaai/opendbc
Quite old and for Wundows, but a lot of code showing how to use a lot of CAN interface boxes is at https://github.com/rbei-etas/busmaster/tree/master/Sources/B...
There's rules of thumb about never overloading a CAN bus beyond, say, 50% utilization. That's because systems with poor prioritization management tend to start falling over around there. With a well thought out scheme, it's possible to push a CAN bus fairly close to 100% utilization. I built several safety critical systems that pushed 80% utilization on average. At that level, you really need to rely on redundancy rather than simple robustness, though. A CAN bus running at 80% falls over very hard when you have a flaky physical connection somewhere.