Renderer performance improvements for Namco System 23

[MooglyGuy]

This streamlines some of the hotter functions in the System 23 rendering code, mainly:

• Templatizing the scanline-rendering functions
• Eliminating the use of rgbutil.h to remove overheard of unnecessary std::clamp calls and alpha-channel processing (which S23 lacks)
• Pre-decoding some aspects of the texel ROMs
• Making render_scanline and render_sprite_scanline member functions of namcos23_renderer rather than static functions.

Overall, this takes Panic Park from around 92-94% for an -str 90 run (with the 'Skip POST' DIP set to On), to around 102-104% on my AMD Ryzen 9 3950X.