[SPIKE] Investigate using JNA Direct Mapping where possible

[skeet70]

JNA is sometimes 40+ times slower than JNI. JNA Direct Mapping brings that performance difference way down. Investigate if we can make use of it, and if so how much performance benefit it may bring. uniffi benchmark fixtures will be a useful place to start for measurement.

[skeet70]

JNR could also be an alternative, benchmarks. There's also project panama, but that seems much more experimental, even though it was released in JDK 19. If we used that we'd be hard locked into newer Java versions than I imagine most people are using.