perf: cache phasor sample interval

[txbrown]

Summary

• Cache the phasor sample interval (1 / sr) when PhasorNode is constructed so each tick avoids spelling the reciprocal division in the hot path
• Preserve existing phasor and sphasor output behavior, including sync reset handling
• Add native CTest coverage plus a focused phasor benchmark target so the optimization is reproducible

Part of #8.

Performance evidence

Measured locally on macOS with AppleClang, Release (-O3/NDEBUG) native builds. The benchmark runs 512-sample blocks and reports median throughput across repeated runs.

Case	main median samples/sec	PR median samples/sec	Delta
phasor	363.8M	364.6M	+0.2%
sphasor	299.5M	297.2M	-0.8%

Result: this benchmark does not prove a stable material speedup in optimized builds. The likely explanation is that the compiler can already hoist or strength-reduce the invariant reciprocal in this tight benchmark. The PR still makes the intended invariant explicit and adds a benchmark target to keep future quick-perf work measurable.

Reproduction commands:

cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=./out/ ../..
cmake --build . --target runtime_benchmarks --config Release -j 8
./runtime/runtime_benchmarks 512 500000 7

Test plan

• cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=./out/ ../..
• cmake --build . --config Release -j 8
• ctest --output-on-failure
• ./build/native/runtime/runtime_benchmarks 512 500000 7