5*1.8 = 9V. 1.8 V is your forward voltage right?
You'll struggle to find an op-amp which will supply that much current, you'll need to use a transistor at the output to supply the current, or employ a current mirror with transistors with differing current gains to multiply the driving current to a sufficient level.
You'll also need to configure it to supply a voltage which is in excess of the combined forward voltages (current source if driving with the op-amp output or a 9V or higher dc supply if using a current mirror to drive them). Both fairly straightforward circuits.
I'm unsure of what you mean here,
"Then I'd use simple peak voltage buffer and voltage feedback to automatically adjust gain so that output peak will stay at 9V. What do you guys think?"
Are you intending to have the diodes turn off every half cycle? Is this necessary? If it is it's probably worth switching the current to a dummy load so that your supply isn't having to jump from almost no current to 400mA every cycle, (a comparator, a logic gate or two, a power resistor and a load switch would do). I'd try and keep the diodes on if the purity of your sinusoid is important. How are you detecting the signal, different methods of detecting modulated signals are more forgiving than others- if you're using a lock-in amplifier I wouldn't stress too much about a bit of distortion.
I'm a touch unsure of what you're asking here too.
"My IRs forward current is a non-linear function of forward voltage (1V will be threshold), however I'd expect some tolerances when driving them. In this case wouldn't it be harder to drive them equally compared to having a current source with fixed max 1.8 Vf? Will they be equal in series?"
I wouldn't expect the forward voltage to vary that much, this should be covered in the data sheet though, so the only variation in brightness will be process/manufacturing tolerances, if you drive them in parallel you'll still have this mismatch plus any mismatch in the parallel legs (however you drive them). I usually use 2 or 3 in series with a number of parallel legs in the drivers I build (the LEDs I use have forward voltages of 4V typically so more than a few in series and voltages start to get higher than is sensible for my application.