Why we picked register-based bytecode over stack.

We looked hard at stack bytecode. We picked registers. Here is why, and why it cost us a contributor.

Stack bytecode is simpler to generate and simpler to verify. Register bytecode has a shorter prologue, fewer memory round-trips, and better behavior under JIT. For Compose-heavy code with many implicit parameters, the per-call overhead of a stack machine showed up as measurable cold-start regression.