Proper 18650's tend to be fairly well protected with integrated PTC and CID devices. For demonstration, I have tried:
* Dead-shorting 18650's,
* Charging 18650 cells from a 30V, 10A lab supply, for hours -
and haven't been able to cause fire, explosion, smoke, or significant venting of electrolyte. Sony cell on a 30V,10A supply worked as a hysteretic PTC thermostat controlled heater, stabilized at around 100 degC IIRC.
Manufacturers often specify such test cases and guarantee* no fire. (* or, at least, that "no fire" is a typical result.)
For reliable demonstration, I'd suggest sourcing some low-quality cells (think about UltraFire etc. brands), and testing thoroughly before the big day.
Physical nail penetration might be the best, if it needs to demonstrate a typical cause. Otherwise, +1 for the in-battery cartridge heater! PTCs, CIDs etc. can protect against internal heat generation over the thermal runaway onset temperature, but using an external source, there's no limit. You should see things start going after your cells reach about 160 degC. (For LFP, that would be higher.)
If it needs to be external and electrical, long-term overcharging (or forcing negative voltage) would have best chances. Shorting a cell is almost a guaranteed demonstration failure - only the very crappiest cells would cause a fire during an external short.
Or, ramp up the voltage enough to arc through the PTC, CID and even the shutdown separator - a few hundred volts, maybe?
It sucks trying to demonstrate a li-ion fire. The chemistry itself is volatile, but they have significant load of safety features; they need to have, many BMS ICs on the market have a long tradition of having some dangerous failure modes, so the cell manufacturers have learned their lessons, their safety process is not completely relying on the pack designer staying within ratings. Yet, abusing a cell definitely increases the fire risk (which is why we do see fires related to, for example, overcharging after the management system failed), but not enough to provide a robust way to produce demonstrations easily (unless you are able to source a consistently very crappy cell).