Optimizing N163 in NSF driver

[HeeminTV]

This PR fixes the current driver with N163 enabled exceeding a frame (single 60Hz tick) by optimizing the waveform copying routine.

Changes in this PR:

• Replaces addition loop to the actual multiplication (This worked before 0CC, which had table-per-instrument limit of 16, but as it's now 64, the usage of CPU cycles in this loop became massive)
• Slighly unrolls the loop for writing to $4800

[Gumball2415]

changing the target to main branch

[Gumball2415]

lgtm, though i'm a bit unsure about the byte order edit in the nsf driver build script

[HeeminTV]

I had no way to make it work properly other than adding new arguments there. It simply didn't work with both Python 2 and Python 3 (which is what CONTRIBUTING.md describes) without inserting updated arguments

[Gumball2415]

CONTRIBUTING.md specifically mentions Python 3.10 and above

• For miscellaneous custom build scripts:
  - Python 3.10+

however, Python docs mentions that the default argument for byteorder was not added until 3.11.

Changed in version 3.11: Added default argument value for byteorder.

[HeeminTV]

Seems like it's CONTRIBUTING.md that should be updated then

[Gumball2415]

changelog:

• Replace addition loop to the actual multiplication (@HeeminTV Optimizing N163 in NSF driver #400)
  • This worked before 0CC, which had table-per-instrument limit of 16, but as it's now 64, the usage of CPU cycles in this loop became massive
• Slighly unroll the loop for writing to $4800 (@HeeminTV Optimizing N163 in NSF driver #400)
• Specify byteorder argument in build_engine.py for Python 3.10 compatibility (@HeeminTV Optimizing N163 in NSF driver #400)