Reduce input delay by 1 frame and audio timing fix

[ghost]

• This PR adjusts timing for input, video and sound. This reduces input delay by reading and pre-process input before the main loop (CPULoop) this way makes the data already available during the loop instead of waiting for vblank to read the input.

• Sound timings and video timings are also adjust, including the number of clock ticks required so that CPULoop will have enough ticks reach finish till vblank which takes about 280896 cycles. This makes it possible that video and sound can both be processed during vblank period which then ends the loop.

#332

[ZachBacon]

I'll approve these changes, but since the CI is acting up on the mingw side of things, I'm not 100%. But you've done good work so I'm more than comfortable to accept this.