Well, if you are using ethanol then to get 3% vapor you'd need it to be at about 10C. Super-cooling is going to drop the ethanol vapor pressure well below the LEL, even though it will be over saturated. Lower temperatures will also raise the LEL.
Liquid ethanol has a flashpoint of 12.8C.

Realistically you don't need to worry.

If you do still want to inert your chamber, I wouldn't bother calculating gas ratios. Just do a CO2 or N2 flood, then add your alcohol.

And 30kV/cm is for perfectly uniform conductors on either side of a gap. 

And thats for dry air at sea level. When the humidity or altitude get up a bit then you could be as low as 10kV/cm.

There are lots of design notes out on the internet. eg https://ez.analog.com/cfs-file/__key/telligent-evolution-components-attachments/00-330-01-00-00-06-34-74/FlybackSnubberDesign.pdf
Start there.

You can also go a long way with SPICE simulation for these things. Characterizing your transformer will be the most difficult, but so long as you stay away from saturation, a linear model will do OK.

Its lacking a clamp or snubber  on the drain.  This means that any energy not coupled out the secondary gets dumped in the MOSFET, in the most nasty way possible: by overvolting the drain.
Example of a flyback clamp:
https://www.ti.com/ods/images/SSZTCV1/GUID-5300FB37-4574-4023-BC85-9BCA1E7D92CF-low.jpg
Note: these things need to be designed, not just hacked together.

Have a look at flyback circuits based on the old favorite UC3845.  This is a current-mode PWM controller - useful as you get direct control over the primary current.

How about two lengths of copper pipe mounted in a block of insulator (wood? plastic?) so that they form an 'X', but spaced apart so a spark gap is formed at the middle of the 'X'. Viewed from the front, one arm of the 'X' would be behind the other. 
The long arms of the copper pipe 'X' should act as a heatsink.  If you can seal the bottom of the pipes too, then you could fill them with water, which would also remove heat by boiling.

Best I can come up with using minimal materials and time.

  I found an IEEE paper(https://ieeexplore.ieee.org/document/4181239) that estimates the resistance of a low voltage arc (anything up 600V or so) as 40/I^0.85

I'm looking for a source for a 20uF 5kV pulse capacitor for use in an electrical surge tester. Does anyone have a lead on a reputable manufacturer? (not interested in used parts)

Perhaps your chosen triode connection is not the best. Maybe try Screen (g2) to anode, and supressor (g3)  to cathode. As suggested here https://www.audiodesignguide.com/New2A3/ETF06TS.pdf
The fact that you know you have shipping damage does not inspire confidence. If the tube got whacked hard enough to short a grid to the plate then lots of other things may be bent out of shape and affect tube performance.

Inductance will be your limiting factor.  A mere 1uH would limit your current to around 30,000 A peak. 
If you ever did achieve 1.5million amps the force pushing your cables apart would be on the order of 50 million newtons per metre.

If you do 'roll-your-own' (in both senses of the phrase) then using multiple thin dielectric sheets is better than using a single thick sheet. The rationale for this is that all plastic sheet will have defects, but with multiple thin sheets these defects are unlikely to line up, and so the multi-sheet stack has better performance.

Then again, you'd need ~0.06" (1.5mm)  total thickness in Mylar, so that will be quite a few layers.

Alternate would be something like 100mm foil discs with 3mm PMMA,PET or PC sheet - that would net you ~80pf, so thats only 5 metal layers to hit ~300pf. You could hot-press the stack to seal the edges and get rid of air.
A 200mm dia foil would get you 300pf in one layer. Then you can take your lead out from the middle of the disc, on each side.
I think I've seen PET sheet thats thinner than 3mm, so that would also be a option.
If it must be cylindrical you could use PMMA tube, foil on the outside, and either metal tube on the inside, or possibly salt water?