Let's work out some of the math here; the details of your build and your DM are going to influence how these numbers work out. Some assumptions I am making based on what you have stated.

Assumptions

1. You said DM is letting you use flanking rules so I am assuming you are getting advantage on each attack so I am going to state that you are getting advantage on your rolls 75% of the time due to other skills and tactics you could be doing.
2. I am assuming that your attack stat at this point is a 20 in any of the builds. I am ignoring barbarians right now who can get more then a 20 in a stat also I am ignoring belts or other items that can raise your stat above a 20.
3. Finally we are going to ignore the Sun Blade because getting a hold of a specific magic item should not be a requirement for a build.
4. This also assumes that your DM lets you roll your X number of attacks and decide to apply the sneak attack to the critical attack. The once per term requirement is very vague on this and should be clarified.

The Math

My first thought is that the thief build is going to be really top heavy; you are trying to get as many dice rolls as possible because you want that natural 20 critical to double the number of dice you are rolling for damage. I did a little look at monsters vs stats and I am trying to figure out a good threshold for when a die roll will hit and I feel like 13 on the die should hit based on average AC of monsters at that level. So 13 for die roll, 2 for weapon bonus, 6 for proficiency and 5 for stat. Now one could argue these numbers one way or another but I am going to be using them as the base because I need something to compare apples to apples.

So 13 + 2 + 6 + 5 = 26 for target AC to hit or 35% chance to hit and a 5% chance to crit. I used http://andrewgelman.com/2014/07/12/dnd-5e-advantage-disadvantage-probability/ as a reference for determining advantage so that translates to a 63.9% chance to hit and a 9.8% to critical. That makes it so 54.1% of the time you are going to do normal damage and 9.8% time you are going to do double damage. I basically am saying 2/3 of the time you are going to hit with your attack, really anything that gives you 2 attacks is going to let you get a sneak attack in during that round so long as you are within 5 feet of your target. This is also a pretty high number, some could argue that 10 or above would be a more realistic number then you are talking almost an 80% chance to hit with advantage.

Since you decided to dip into fighter instead of go 20 rogue you don't get Stroke of Luck but that would only be a factor depending on how much resting vs combat you would do. To ease math on calculating average damage I am assuming your weapon damage for rogue is a D6. If you are getting a D8 or better this just slightly makes your numbers better it just helps me to do my math faster.

Critical Hit - 18d6 + 2d6 + 5 + 2 = 77 Avg

Normal Hit - 9d6 + 1d6 + 5 + 2 = 42 Avg

GWF Critical Hit - 18d6 + 2d6 + 5 + 2 = 85 Avg

GWF Normal Hit - 9d6 + 1d6 + 5 + 2 = 46 Avg

The math was kind of surprising because of the limited reroll only on 1 or 2 and you could get a 1 after you rerolled a 2 but on average on 10d6 rolling a 1 or 2 and getting 4 extra damage on average.

Polearm Master

After looking at the polearm master the numbers are basically the same for attack and damage except that you get another attack that you will get your critical damage against. Even though the base dice is a d4 that extra attack is what gets you the key advantage — do you want that extra damage from the reroll or the chance to attack again and be able to sneak attack? Frankly I'd rather take the 63.9% to do my sneak attack dice again then the extra 4 damage on average.

Crossbow Expert

This basically follows the same rules as Polearm master: the feat is letting you get an extra attack with the a crossbow and I still think it is better then Polearm master because being able to attack at range is crucial; however, that depends on your flavor and what you like to do. However where this starts to break down is flanking and how your GM rules on flanking with a crossbow. If your GM rules you can't with a range weapon then you are stuck with polearm.

Bottom Line

Bottom line your GM house rule about flanking is getting you the most bang for your buck here because you want to maximize that rule. The winner between all 3 is really hard to decide, and also depends on how your GM house rules critical hits and determining which one was a sneak attack. If you get to pick the extra attack is going to be always worth it. If you can't pick which attack to apply sneak attack damage to then they are really almost a wash because your extra critical chance vs the on average higher damage from your rolling is going to end up being almost a wash in the long term at high level.

I wrote a python script to calculate answers accurate to several decimal places. I created a Moonblade object, and added a specified number of runes just as the rules suggest (with the one alteration you made that you always re-roll non-stackable properties). It then calculates the expected damage from this moonblade taking into account all bonuses, bonus dice, critical hits, creatures, etc.... I made a couple of assumptions beyond what you have explicitly stated.

1. I assumed that the creature on the receiving end has a uniformly random chance of being any of the 14 types. This is a bad assumption for most games (how often do you see a plant or an ooze compared to a humanoid?), but absent more data, I cannot do better.
2. I assumed that the roll to hit was greater than 1. Though you did say to assume that the hit has been made, in order to calculate the chance of a hit being critical, we need to know the chance of the d20 showing a sufficiently high number given that we know that it is high enough to hit in the first place. The chance of a random d20 roll being a 20 is 1/20, but the chance of it being 20 given that it had to be able to hit an AC 18 creature could be significantly higher. Since we do not have an AC to measure against, the best we can do is say that we know the roll must be higher than 1 (as a natural 1 never hits).
3. I assumed that the user never uses the property of the defender whereby he can transfer some of the sword's hit bonus to his AC.
4. I assumed the user always uses one hand.
5. I assumed the user has a damage modifier of 5 because you would expect this of the high-level character wielding the moonblade.

With these assumptions, I get that with 99% confidence, the true expected damage values given \$n\$ runes are these, \$\pm 0.05\$.

\begin{array}{ll} 0\text{ runes}&9.737\\ 1\text{ runes}&10.244\\ 2\text{ runes}&10.752\\ 3\text{ runes}&11.252\\ 4\text{ runes}&11.727\\ 5\text{ runes}&12.166\\ 6\text{ runes}&12.557\\ 7\text{ runes}&12.899\\ 8\text{ runes}&13.196\\ 9\text{ runes}&13.459\\ \end{array}

This plot demonstrates the almost linear relationship.

For the first 10 or so runes, the expected damage is about \$9.912 + 0.421 n\$. However, the linearity drops off at about \$n=10\$. This is because we expect to have reached a bonus of +3 by this point.

If we extend to a truly absurd number of runes, we see a transition to another linear behavior. I think this is because the dominant factor in the damage is the steadily growing number of d6's added to the roll. Because all other runes eventually stop stacking, in the end, all we can do is add another d6 for each rune, meaning we either add 1d6 to the damage, or add 1d6 to the damage for a particular type of creature. After adjusting to account for critical hits, this averages out to about 2.07 damage per rune, which is what we expect the slope to eventually be. Therefore, for large numbers of runes, the expected damage is calculated as \$\overline{d}\approx-93.24 + 2.07 n\$