You'd be better with a current mirror instead of R1 (then you don't have to think about what value to make it).
You might be better with a MOSFET instead of the Darlington, but a Darlington may be better than a single transistor, as you have much more Gm.
If the load is single ended (i.e. doesn't have to go both negative and positive) then the MOSFET could drive the load, and you don't need T5 and T6 and even to think about bias.

I'm not sure the typical advantages of the mirror would apply for the application being the I/O is only DC. But great idea on the MOSFET, I had planned to look into a CFP or something today. I'm not opposed to reducing component count and/or making this thing better.

The load is tricky and simple at the time, so I'm not sure if we can dispense with the push-pull output or not. It's literally 2 series diodes (anode to output) to keep the output DC local into a node with AC and a 220k resistor essentially tied to the input. So the output of the opamp needs to clip or absorb some AC but only at extreme negative end of the waveform. I'll draw it up and post today for discussion.

 

Ok, so new diodes in and the initial tests look good. Attached is a schematic with the I/O circuitry shown and the voltages I measured. I haven't yet tested with a high -Vin or tried to adjust the feedback yet (currently set in the middle).