Startup and exit take 0.7 ms longer than stock hxcpp 4.3.53 #234

A program that does nothing takes 4.0 ms to start and exit with libhxcpp, against 3.3 ms with stock hxcpp 4.3.53 (median of 20 runs each, alternating, measured by the benchmark driver around the noop command). That is about 0.7 ms, or 21%, more per process.

It matters for command-line tools, build steps and test runners that start many short processes.

Not known yet: where the time goes. Candidates to measure, not confirmed: static initialisation added since 4.3 (registration tables, telemetry or GC setup), the initial heap reservation, thread or TLS setup, and anything done at exit (finalizers, flushing, GC teardown). Comparing an ETW or perf profile of Main.exe noop for both runtimes should show it.

Measured on perf/benchmark-suite at bf583e5: Windows, Ryzen 9 9950X, MSVC 19.44, 64-bit, Haxe 4.3.7.

Reproduce: the "Startup and exit" row of the driver's output.

cd bench
haxe --run Run --only map_int

(--only still measures startup first; any benchmark name will do.)

Done when startup and exit are at least level with stock on Windows and Linux.