1) You should never use digital control for a dc/dc controller (i.e. using a microcontroller's PWM and ADC in a control loop)
2) DC/DC converters are generally inherently so safe that it is not worth the time, effort and parts to add failsafe protection
To part 1: a dc/dc converter can be proven to be stable over a sense and duty cycle range when dealing with (close to) analytical control loops, e.g. linear feedback, opamps, passive compensation networks feeding into a PWM generator. You can say it's stable because it has a definite frequency response: bode diagram, nyquist diagram. With purely digital control you lose this kind of behaviour, because you need to implement it into software. To make a purely digital dc/dc controller you need to:
- Sense the feedback voltage (e.g. for LEDs: voltage over a current sense resistor)
- Combine this with information about earlier states and do some matrix multiplication to simulate a digital compensation filter -> this is now the output of your 'digital error amplifier'
- Feed this into the PWM generator and make sure it changes periods glitch-free
The amount of time it takes a reasonable-cost microcontroller to process this information - heck, even the time to do one SAR ADC cycle - severely limits the bandwidth and propagation delay of this kind of a control loop. For example, on an Atmel XMEGA with 2MSPS ADC, you would have a propagation delay of 3.5µs for the ADC (7 ADC cycles) and you would need to do at least n^2 multiplications for an n-element filter, with 4 elements being the absolute minimum to be able to implement a decent filter. Multiplications take 2 CPU cycles, so at 32MHz that's about 1µs calculation time. 4.5µs propagation delay means your minimum duty cycle times frequency should be more than about 5-10x the propagation delay to minimize phase effects. i.e. 23-45µs. Even if you were certain that your converter would always be running at optimal duty cycle for this control scheme (~50%), you would still be limited to about 10kHz PWM frequency. And this all is the absolute best case for such an application.
You will need a proper control loop and for any kind of decent switching regulator that will need to have an GBW of a couple of MHz, propagation delays of no more than a few ns (which can often be compensated with feedforward capacitors if you're really eager). This is not attainable with a purely digital design, and I would advise against even trying to do this with a microcontroller. DSPs and FPGAs can be used, as well as a few microcontrollers with built-in power management peripherals. For your application, I would advise to use an integrated buck LED driver or a linear LED driver. Switched capacitor is not going to work here - it will have worse flicker performance and worse efficiency than linear.
Now to part 2: If designed correctly (which usually just means: you read the datasheet thoroughly), the chance of failure of a buck converter under normal operating conditions is slim to none. The control loops are generally well-compensated and tightly coupled so there is no chance of any event (pulses, EMI, etc.) causing an overcurrent event.
However, the keyword here is 'under normal operating conditions'. The best way to shield yourself against problems is to make sure you maintain these conditions. Use a reverse bias diode and (resettable) fuse to protect against accidental reverse connections, use an MOV or TVS and bulk input capacitor to protect against insertion voltage spikes caused by the inductance of the power wires. Use differential and common mode LC filters to decouple the board from excessive noise on the power line. Depending on how hazardous your particular power source is you may choose to leave out all or part of these protections. Using them all is complete belt&braces; not necessary for 99.9% of applications but if you're feeling exceptionally paranoid feel free to put them all in.