Use JNI to speed up performance of interfacing with simulators

[edwardcwang]

See chipsalliance/chisel#798

[grebe]

See also my chisel-testers WiP PR

[ekiwi]

We are using JNA now with Verilator. With JNI the problem is that it is tricky to load more than one simulation at a time into the JVM. There could be a way to further speed things up by essentially implementing our own JNA-lite. If anyone is interested in taking that on, please talk to me.