There is a better way to do single supplies, ie an amp derived floating ground. Single supply amps have to be wired carefully to avoid sharing 'ground' with the OP supply. I would put the output cap after the Zobel network, in case the 2200uF cap has high ESR and/or inductance. Yes, the drivers seriously need to be bigger, 1 Amp or larger and 5 or 10 Watts. You want all other poles besides the (VAS) dominant pole to be as high as possible. I have used a RC+RC network for 50 Ohms ~100MHz. The various parts are more-or-less stand alone, so if you want to skip one or more, that should not be a. This web page attempts to demystify the process. This is referred to as the frequency domain behavior of a system. ~Zero at high frequencies is the ideal, but you can't do that on the negative input. Bode plots are a very useful way to represent the gain and phase of a system as a function of frequency. 1K on the feedback divider is better than 2k5 but there is no reason to do the same to the positive input. The VAS base resistor should be 2x+ the ccs resistor, but a current mirror would be better. Yes, the IPS current is high but I left it as is. You can cross-couple a CFP with a small cap(+R), but a larger cap creates PSRR problems. I show the driver current as about 12mA in simulation (and 0.65/47=13.8mA), but 100 Ohms on the OP base is marginal for shoot through and turn-off time, 80% of BJT amp failures. Yes, the input should be decoupled/filtered.