The one detail that I don't think has been mentioned is that this phenomenon does not apply to SSRs because they have "zero-crossing" circuitry
I'm fairly certain we understand the underlying process of electrical arcing as applied to when a switch is used to break a circuit, so it wouldn't be called a phenomenon.
Also, the reason an SSR does not arc is not because it switches at the zero point, it is because it is solid state, there is no mechanical contact to arc and no air gap.
There are SSR's that do not switch at zero point, used for example in phase angle control, turning on at some angle other than 0 or 180 degrees through a power cycle and they do not arc.
If the SSR is adequately rated, it should be fine to have it permanently in circuit as in the remote possibility that it fails (and fails "ON"), the worst that can happen is the heater runs up the power bill.
Paulcs:
For a water tank, using one SSR on one electrical leg is perfectly adequate for control purposes. I would probably use it in SERIES with the existing thermostat. Set the Arduino to a middle temperature, such as 115F, and set the water heater thermostat to a higher temp such as 135F. The Arduino does the day-to-day control, and the thermostat does a "kill switch" at a higher temperature for safer operation.
This comment seems to be entirely "out-field" for the present discussion - I cannot see the relevance.
Obviously, the SSR (or contactor) will be external to the heater, will not interfere with the normal heater operation in any way, all thermostats and safety devices are untouched.
If the current required to control the SSR is (significantly) less than 30 mA at 5V, then it might be practical to connect the SSR directly to the Arduino, though you would not get the full 5V at this current. It would probably be preferable to switch it with a NPN transistor via a 1k resistor to the base, the emitter being grounded and the collector going to the negative control input of the SSR, the positive of which is connected to the 9V power supply which feeds the Arduino.
I understand that the way I can command the SSR to close the circuit is by feeding it with 5V-32v current. /snip
I'll have the 3 phase SSR interrupting the three phase cable that goes from the 380V mains to the water heater and at the other end of the SSR I would use a transformer that outputs anything from 6 to 30V).
Well, which is it, is it 5-32 volt or 6-30 volt control signal?
In my experience, AC type SSR's typically have a wide 3-32Vdc input control voltage, not 5 volts or 6 volts as the lower end value.
I have a Opto-22 240DS25 SSR connected directly to my Arduino Mega, working fine.
It switches a resistive heating load element as the shunt load from my hydro turbine system.
The Arduino controls pin D2 via a PID loop control and switches the SSR in slow PWM mode when needing to shunt the extra energy. So, when in shunt mode, is is switching the SSR around 5 times per second, and has done this for around 2 or 3 years now without problem.
Again, be aware that you are dealing with mains power potential and as usual, I would recommend you use the services of a suitably skilled and qualified person to do the actual work of design and installation.