Your default dwell times are wayyyyy too short.
On high capacity systems, train dwell time becomes the limiting factor on passenger capacity. 30 seconds is generally the minimum possible dwell time a system can manage, 20 seconds might be possible during much lower demand periods. But you’re unlikely to ever achieve better than that.
The London Victoria Line, which runs with 90 second headways at peak, achieves at best 24 second dwell times in central section, but 30-40 seconds is more realistic for most stations.
Don’t forget, dwell time includes more than just passengers getting off and on. It has to include time to open the doors, close the doors (including a 2-3 second visual and audible warning!), perform needed safety checks, and eventually pull away. Those operational components the sandwich the core “people getting on and off” bit of station stops add up to a non-trivial number of seconds.
https://www.researchgate.net/figure/Dwell-Time-and-Passenger...
As a minor note, the NYC subway uses a system of fix automatic blocks, not a moving block system, which is what your simulation appears to use. Moving block systems are dramatically more efficient than fixed block systems. But I have no idea how you would get hold of accurate block locations for the NYC subway.