Thanks Dennis and Neil.

I pulled out the AFM and bench-tested it both upside down and
right-side-up with a power supply and oscilloscope and could not get
it to misbehave.  Nice smooth waveforms as I moved the vane by hand.
I could also see the action of the capacitor when I went faster.
Hmmmm.....

Regarding O2 sensor issues, I'm pretty sure I'm OK there.  I have a
dash air-fuel gauge (basically a voltmeter) hooked up to the O2 sensor
so I can see what's going on.  The hesitation happens also when the
engine is cold and the ECU is still ignoring the O2 sensor and using
only the AFM.  At that time, during hesitation, the mixture always
goes lean.  It's unclear to me whether the hesitation is caused by the
lean mixture, or whether the mixture reads lean because there's air
(and fuel) in the exhaust from a misfire.

When the ECU is closed loop and hesitation occurs, the mixture
temporarily slams rich and/or lean.  This looks different than the
normal closed-loop back-and-forth sweep.  I assume the rich is the ECU
trying to compensate for the lean condition.

I spoke to the AFM vendor yesterday and he suggested I check for
vacuum leaks.  I had hoped to be in the desert now, but it looks like
I have to get back to basics (unmetered air leaks, ignition, etc...).
Here is my plan for Saturday, in rough order...suggestions/additions
welcome.

Check/inspect intake piping and vacuum lines
Check/retorque intake manifold to head (this is an I-4), throttle-body, etc.
Vacuum test brake booster system from manifold connection
Vacuum test fuel pressure regulator from manifold connection
Possibly try running with vacuum lines/PCV plugged
Inspect distributor/ignition wires
Inspect plugs
check compression
Tune-up (timing, idle speed, Idle CO)

Two bonus questions for any Digifant II pros:

1. For idle CO, I was not able to find any reasonable home-mechanic
options for an exhaust gas analyzer (suggestions?).  From what I
gather, the CO mixture spec corresponds to just rich of
stoichiometric, so I plan to use the O2 sensor output to set this
(i.e., start lean, make mixture richer just until the sensor output
flips rich).  Does this seem reasonable?

2. Does Digifant II have any long-term learning/adaptation to the fuel
map?  When I replaced the AFM in 2018 and did the idle CO as above,
the engine first ran a little rough (i.e, blip the throttle for a
shift and get a misfire), but then got smooth over the next few days
or so.

Thanks as always,

Alex