I assembled one of the second pass 30A prototypes using different FET switches than the first pass -- substituting a  Vishay SiSS28DN for the IRLHM620. The Vishay part claimed slightly lower RDSon (1.6mΩ vs. 1.8mΩ @ VGS = 4.5V) and a higher max VDS of 25V instead of 20V. I thought, erroneously,  that the higher voltage rating might improve the survivability of input transients when batteries are plugged into the input. The addition of R1 between the input and the LDO solves the problem by eliminating the capacitor (C2) that caused the ringing so a higher voltage rated FET is not required to support 2S-4S LiPo batteries.

I was able to use a different FET because the 3.3mm x 3.3mm DFL-8 package is apparently a standard. 

Functionality:

The switch no longer is sensitive to disconnection by high currents. I ran more than 35A through it and it remained closed. 

It is now tolerant of the input transient caused by plugging a 4S LiPo battery into the inputs -- there is no measurable ringing. 

Those were the two major items that had to be fixed on this pass.

The new Hall-effect device trips at 45 Gauss and requires a stronger magnet. I ordered some 6mm x 10 mm N52 magnets and now the switch trips when the magnet is about 5/8 inch (16mm) when directly over the Hall-effect device. Since the Hall-effect device is active high, the switch activates and deactivates when the magnet is applied, rather than removed, which makes this measurement simpler (and the operation of the switch more straightforward).

Measurements:

Off-state current :

5.3µA @ Vin = 6V

5.7µA @ Vin = 21V

On-state Current: 7.2mA @ Vin = 16.8V

Switch resistance: 

1.22mΩ @ 8.2A

1.23mΩ @ 14A

1.3mΩ @ 33A

1.3mΩ @ 35A (45mV across switch)

EDIT 2018-03-29:

The switch resistance measured above was taken from the two pads on the board. While testing the other 5 boards I noticed that the resistance was 25% lower when measured from the copper landing that I had placed near the sources of M1 and M2. That's significantly lower, which I did not expect. I connected up two more boards, using the landing pad for the B- lead. Here's what I measured with that configuration:

1.0mΩ @ 1A for the board using SS28 FETs.

1.2mΩ @ 1A for the board using IRLHM620 FETs.

In the end, I did get a pretty good improvement in switch resistance from the SS28 devices.

Memory hold time is between 30 seconds and 45 seconds. The lower off-state current drain has caused this to lengthen. C6 should be reduced to get this back into the 15-20 second range.

Conclusions:

Meets both functional and parametric requirements. No circuit changes necessary. I'm going to build a total of 6 units -- 3 or 4 for my use and the rest will be sold. I'll report back on how they perform in my electric powered planes.