3.4Vrms is close enough to 3.2Vrms; not usually an issue.
30mVrms superimposed on the -20VDC isn't an issue either.
VFDs are fairly robust. Just don't grossly over-voltage the filament or segment/grid pins from recommended values.
AC is preferred between filament pins although you can drive them with DC but then you'll get a noticeable gradient especially on long displays.
The ZD2 zener appears to be doing its job seeing that you measure -20VDC. So do the diodes given the low 30mVrms; possibly due to leakage so unavoidable yet inconsequential (3 orders of magnitude less).
What's interesting [to me] is that there is NO Ek bias at all. By design!
The filament winding, instead of having a 3rd center-tap wire, derives it from a simple R7+R8 divider.
Usually, you'll have another 2-4V zener between center of R7+R8 and -Vee.
This would make the "center-tap" of the filament a zener drop HIGHER than -Vee.
ie. -Vee(-20VDC)---2-4Vzener---filamentCT------grid/segment---PFET---Vdd(+5VDC)
And therefore, the ends of the display where the AC swing is the highest (1/2*3.4Vrms) is still kept completely above -Vee.
Q: Why keep the filament (at the ends) above -Vee?
A: Because when a grid or segment is turned-off by an HT16512 output an internal resistor pulls it to -Vee which will be BELOW any filament potential guaranteeing it goes completely dark.
So if you are seeing lit segments/grids at the ends, ergo, the filament ends must be going below -Vee!
Yeah, but it'll only be 1/2*3.4Vrms below -Vee which isn't really enough to turn-on a grid/segment.
Q: Could there be something wrong with the HT16512 output where the PFET isn't shutting-off completely?
A: No because it's only the ends that are lit but the VFD is multiplexed and a bad output would exhibit the same problem equally across all digits.
Q: But given that there is no Ek bias by design, was this always the case and you just noticed it now?
A: Not likely since seeing only lit ends is definitely a sign of something odd.
Q: Is turning to the lowest brightness level the way to go completely lights-out?
A: Probably since some people are very particular about seeing any lights in a dark room and this is a likely function provided.
So, I think I can conclude that something changed to raise the lowest brightness from just below noticeable to definitely noticeable.
Q1: Could your mains be at the high-end of what the design expects? ie. 125VAC+ vs 120VAC nominal which translates into a higher filament winding voltage.
Q2: Has the combined R7+R8 resistance gotten higher such that more current is directed to the filament?
I can't say for sure that it's just the higher filament voltage plus no Ek bias being enough to turn on the ends (maybe with slightly leaking PFETs contributing too).
But I can confirm that one of my VFDs does begin to light-up at +5V between filament and grid/segment. So maybe.
You could mitigate by adding some resistance to R4 and R3 (from 1R to 5R) and see what happens.
It should lower the overall brightness with the lowest setting going completely dark.