Do anyone know why some fixtures have open board ignitor, while some have enclosed ignitor option? Is it for cost savings or something? I wouldn't think a plastic cap would add much to cost?

A circuit with 5kV output that works in quite dirty environment (dust accumulation, moist air, ...) gotta be encapsulated to prevent eventual isolation breakdown. Its common sense. The only thing that stands in the way of common sense is cost saving

A circuit with 5kV output that works in quite dirty environment (dust accumulation, moist air, ...) gotta be encapsulated to prevent eventual isolation breakdown. Its common sense. The only thing that stands in the way of common sense is cost saving.

That's the problem with common sense, it's not all that common. Particularly amongst bean counters 

Ignitors (maybe except "smart" ones) dont care about surges, since their internal circuitry starts with a big RC filter

Sorry, a lightning surge would laugh at that filter.

In such case there would be nothing left not only from the ignitor, but also from the ballast, as well as the capacitor and even wiring...

The open board may well take all comon (so not the direct lighting strike) overvoltages even when covered by the dust and dripping wet. Its all about surface distances and for that exist quite cheap remedy: Cutouts in the board.
The encapsulation may work, but only until it delaminates. Then it serves as a prevention of the water drying out, so makes the matter way worse.

On potted European ignitors the internal components are before the potting "free flapping in the wind", what is the only thing preventing any surface creepage. So then the most common fault is the fried pulse transformer (its secondary is in series with the lamp, so heated by the arc current).

With sufficient dirt and moisture there are leakage paths forming over the somewhat longer distances with cutouts. They dont make as much distance as a terminal block where the path must form all around and into it, with 15mm+ distances. In a few years cobwebs will build up over any PCB cutouts anyway

And it's not the one off surges from the grid that are the problem. The ignitor itself generates 5kV and even series of 5kV pulses when the lamp is hot restriking or cycling. This is sufficient to start arc tracking, that then can continue at low voltage

Also, the component that really does not take surges well is the PF capacitor, placed directly across the input. They tend to self heal down to useless capacities over the years, and sometimes fail spectacularly too

Unless the voltage surge was less than what would destroy the non solid state components. As you probably know when a surge enters a house the cordless phones and electronic devices are fried way before the lamps, motors, or anything else electrical that does not contain solid state circuitry. It isn't any different on the street lights as your P/C, Ignitor are the first to go.

But the thing is, ignitor circuit inherently contain a low pass filter, so the only component really exposed to the high voltage is the resistor, nothing else. And because the resistor has to be at least 5W rated just for thermal reasons (mainly when the lamp failed to strike), we are talking about way more than a 15kV capable device. And that is way above what the ballast wire insulation can usually handle.
Of couse, assuming well designed board (proper cutouts, so with sufficient creepage distances of 20mm and above)

With sufficient dirt and moisture there are leakage paths forming over the somewhat longer distances with cutouts. They dont make as much distance as a terminal block where the path must form all around and into it, with 15mm+ distances. In a few years cobwebs will build up over any PCB cutouts anyway

Well, with just the 4 components (resistor, choke, capacitor and sidac) surface distances way above an inch for the exposed components (mainly the resistor, inductor and mounting places) and given the required component sizes, cutouts yielding surface distances of an inch and above are not that difficult to make. In fact because the resistor and inductor are both larger than that inch, we are talking just separating the connection points from the mounting hardware...

A pole with a pole transformer with higher insulation levels (BIL) will survive a surge more than a street light and the ballast windings in a street light will survive when the solid state components will not.
There are always exceptions due to the erratic paths that a surge can take which are dependent on many factors.

NOTE: Let me clear up on the failures. Failures that I was expanding on were the premature failures and not the EOL of the fixtures or lamps.

The story with the pole is different : On the poles with transformers, there is the main connection of secondary Neutral to Earth. This limits by far the possible common mode surge, so everything that would break isolation towards the luminaire body. Maybe the lowest grid impedance at this point limits also the differential surge, but i doubt it since the transformer secondary inductance is still significant impedance for a fast surge

I dont think melting capacitors in few years old luminaires count as normal EOL ?