Thanks for your input everyone.
ngoma, I didn't do the seal myself last time so I didn't get to see how the old one had worn/failed.
The IP belt was refitted by a local garage after head gasket replacement last year but it doesn't feel excessively tight. About 1/2" deflection available mid-way with moderate pressure.
I don't have a stethoscope yet but I think it must be the shaft seal. (Sound is definitely coming from that end.)
I clamped off the return hose today and applied an airline to the filter feed pipe. Started out real gentle at first - 0.5 bar or less - to see if I could get some fuel to come out the leak. No joy. With the return clamped, the darn thing was air/fuel tight to 3 bar.
Whatever path the air is taking to get in and allow the fuel to drain back, it's a one-way path. Most likely applying positive pressure, as I did today, is causing the shaft seal lip to re-seal against the shaft. Only other reason I can think of for the case to be airtight to positive pressure but leaky enough under vacuum that gravity and a few grams of fuel will pull air in is that one of the pump components is effectively acting as a one-way valve between the fuel inlet union and the leak. Maybe the pressure reg valve - I'll have to remind myself of the exact layout of the pump internals...
It is very odd that there is no sign of air ingress when the engine is running.
I eliminated the stop solenoid by applying power just after the engine had stopped - the air still got in as before.
(Annoyingly thought of this logical test AFTER buying a new solenoid.)
I was going to change the solenoid anyway but it seems to me you have to remove the injector lines to get a spanner on it. Went to do that and found that one of the delivery valve housings was loose in the pump head and so turning the gland nut on the pipe was just loosening the valve housing out of the pump rather than the injector line from the valve. Nightmare!
Of course, it had to be the most tucked away one at the bottom and on the motor side. I don't really understand how it wasn't leaking fuel here or mis-firing. The delivery valve housing was properly waggling about. I was running out of time and the solenoid had been pretty much exonerated anyway so I admitted defeat and simply carefully tightened the whole thing back into the pump. No sign of twisting the pipe so that little problem will have to wait until the pump comes off.
Which it's looking more and more like it's going to have to. I can't afford the dial gauge to time it back up at the moment but I think that's the next step. Depending on the price a seal/bushings kit and LP pump paddles etc., I might even have a go at a rebuild...
Just at the moment, the pump is making me curse a fair bit but I have to remember all the distributors, HT leads, spark plugs, MAF sensors, vacuum lines and ECUs I'm NOT having to deal with thanks to its existence
745TG, were you thinking I could lock the pump with a pin and lock the cam somehow to be able to pop the belt and sprocket and not need a dial gauge?