The NE592 has an equivalent input noise of 10 microVolts RMS in 10MHz bandwidth (per ONNN SEMI datasheet). In one Hertz BW, we divide 10uV/sqrt(10^7) to find 3.16nanoVolt noise density, or Rnoise of 600 ohms.
The MIC920 has 11nanoVolt noise density, or Rnoise of 25,000 Ohms. Thus the MIC920 costs 10dB or more of SignalNoiseRatio, and needs external components to set the gain. But MIC920 has lots more output drive ability.
If you want the NE592, just leave one side open. Be gentle in demanding current to charge a peak-detector thru a diode. The NE592 has internal feedback to prior (internal) stage, implementing a Cherry-Hopper gain-stage for linearity. Capacitive loads upset the delay (the phase) and may produce peaking or oscillation. But such is a risk for any "opamp" driving a peak detector.
You can make an envelope detector out of NPN, base biased barely on with a similar NPN, resistor in emitter to linearize the rectification response, and a R+C collector load to set gain and bandwidth (to track modulation).