PHLEGM v6 NES native hot-path fix - 2026-10-03

Observed runtime symptom
- NES Donkey Kong remained below real time even with PPU/APU diagnostic skips.
- Asynchronous host presentation materially improved throughput, proving host paint
  cost mattered, but CPU/native support dispatch was still host-bound.
- Commodore machines did not exhibit the same failure because their lower CPU clock
  and non-native-raster workload leave substantially more host-time margin.

Changes
1. native_runtime= remains authoritative data. Its normalized identity is now cached
   once as an integer nativeRuntimeKind when a support-chip instance is loaded.
2. Instruction-rate native dispatch (advance, NMI/IRQ probes, CPU register I/O and
   NES/SNES render dispatch) uses that integer cache rather than LCase/Trim/String
   comparisons hundreds of thousands of times per second.
3. NES PCM boundary scheduling now keeps an integer CPU-cycle countdown. Floating
   point division is performed when a 44.1-kHz sample boundary is established (or
   emulation speed changes), rather than for every emulated CPU instruction. The
   APU is still advanced to the same sliced CPU-cycle boundary before sampling.
4. No NES clock, opcode cycle, PPU VBlank/NMI, mapper, or Commodore behavior was
   intentionally changed. Diagnostic PPU/APU Skip semantics remain unchanged.
