Enhance Multiway implementation

[mucaho]

Benefits:

• Limit diagonal movement to same magnitude as horizontal or vertical movement.
• Allow diagonal movement and arbitrary direction movement with correct speed when pressing down multiple keys (may be of interest for isometric)

Caveat:

• Takes an additional 5% of total frame time on top of current approach in worst-case scenario (smashing combination of 8 keys on a blank screen with 1 entity).

Notes:

• See the 8-way example.
• Velocity values handled by Multiway are truncated, so diagonal movement is approximated (can be changed).
• Solves issue described here
• I tried many alternatives to the current implementation, but my benchmarks showed it's not worth it on newest firefox and chrome
  • checking if we move in only one direction / diagonal so we can avoid Math.sqrt calculations, not worth it
  • memoization of input parameters, not worth it
• There is probably a faster approach to handle diagonal movement only (current one is for generic directions)

[mucaho]

I'm still on the fence whether the additional correctness & complexity is needed for Multiway or not (at least for its current uses).
What do you think @starwed? I'd like to hear your input on this.

[starwed]

What's the point of all the e6 stuff?

[mucaho]

Each direction vector is added to the net movement vector when key is pressed.
Each direction vector is substracted from the net movement vector when key is released.
Pressing and then releasing 5 different direction keys should net into 0 movement.
However, due to rounding errors, you don't get 0.
That's why this net movement vector is represented as a number * 10e6.

[mucaho]

I have put some more thought to it, it's just overly complicated.
Who needs more than 8 directions for keyboard control?
Joystick / Gamepad support (with arbitrary directions) is already provided by external component.

[starwed]

Closing, since mucaho ended up feeling it was too complicated, and it's also superseded by #1051