Benchmark and optimize class lookups

When https://github.com/mapbox/mapbox-base-android/issues/14 is closed, we can look into running benchmarks on release builds and measuring the impact of reflection class lookups. https://developer.android.com/jetpack/androidx/releases/benchmark looks like a good tool that we can use.

Optimization ideas, TBD:

• potentially, kotlinf-reflect is able to cut down invoke calls and class lookup times. We'd need to evaluate binary size increase vs performance improvements if those are confirmed
• for all other clients that are not building from source but consuming the pre-built binary instead, we could hardcode the glue classes paths and avoid the reflection overhead

Start by checking the prerequisite issue #14 and the AndroidX Benchmark documentation. Benchmark release builds to measure reflection class lookup performance, then evaluate the binary-size and performance tradeoffs of kotlin-reflect and hardcoded glue-class paths. Done means the alternatives have measured results and a documented optimization direction.