OK Before getting to the question a picture and good news.I've replaced the rotary switch with some toggles. These NKK toggle switches have ~0.25 pF between the pins, which is a bit less than the C&K toggles (~0.38 pF) This allowed me to get the ~1 MHz BW I need (at R_tia= 100k ohm).
So today I wanted to investigate the bootstrap (BS). I built this simple circuitAnd measured the step response and bandwidth (BW).
With no BS, the step response had a 1.2 usec TC and ~140 kHz BW, all in agreement with ~120 pF of PD capacitance and 10 k ohm.
With the opamp BS, Step response had some uglies, ~400n sec (two pole) and ~600 kHz BW. PD capacitance reduced by a factor of four. (Everything I read says I can do better.)
The Jfet bootstrap was less good, TC = 500 nsec (single pole) and ~400 kHz BW.
Finally the npn (2n3904) was nice and snappy, ~160 nsec rise times and 2.6 MHz BW! But I'm stealing a bunch of the photocurrent. (I_col ~3mA, I_b ~ 10 uA) So not practical.
So any ideas? A darlington? How do I pick a better Jfet?