Today I got more specific specifications.
The minimum photodetector size is .3 mm. Larger would be preferred as this would provide a larger target for the laser beam making easier alignment.
The laser's the total irradiant power on the photodetector will range from 10 uW to 5 mW. Linearity must be maintained over the entire range. Of course they would like to be shot noise limited over the entire range, but they recognize this might not be feasible. They are open to having two or more units, each optimized for a narrower range of input power levels. At 40 MHz how practical is it to switch feedback resistors?
The modulation will be roughly 1% of the DC level.
The DC output band will be from 0 Hz to 50 kHz The AC output band will be from 20 kHz to 40 MHz. They do not want bandwidth any higher that 40 MHz for noise suppression reasons. The signal of interest is 40 MHz and below.
I see your point that any source that puts current into the virtual ground to tune out the DC offset will introduce its own noise.