llama : NvAPI performance state change support

[sasha0552]

Related: #8084

Reference implementation

TODO:

• [x] Implement performance state switching functions
• [ ] Place performance state switching calls in a common function before/after inference start/end
  • [ ] Switch only if Pascal GPU(s) present
• [x] Compile only if CUDA enabled
  • [ ] Enable by default if CUDA enabled, otherwise disable
• [ ] Log performance state changes and library loading status
• [ ] Synchronize pstate changes between n instances of llama.cpp on a single GPU

• [ ] Clean up temporary/debug code

• [x] I have read the contributing guidelines
• Self-reported review complexity:
  • [ ] Low
  • [x] Medium
  • [ ] High