feat(kore): ts step interpreter

[paradite]

This PR implements the new step interpreter for TS, which allows full interactivity during the game at each step.

This should allow for easier training in TS.

Major changes:

• added miner bot (MinerBot.ts, miner.py) and do nothing bot (DoNothingBot.ts) in TS.
• added a new mode step in the interpreter, alongside the old run mode. The way to initialize the board is a hack, but it works for now.
• change KoreIO to async as training in tf.js involves async function calls.
• fixed a few bugs in the kore/Board.ts, kore/Direction.ts and kore/Point.ts which is throwing errors when I tried to use board.next() to simulate full games. Might need to add some tests for these fixes?
• export tick function in TS bots to allow them to be imported inside the interpreter directly, so the training code and the agent code can run within the same process.
• added episodeSteps in kore/Configuration.ts.
• made currentPlayerId mutable so that we can rotate currentPlayerId in interpreter to allow agents to tick interactively on the same board.

Out-standing issues:

• The agent behaviour is not exactly the same in TS step interpreter compared to the python one. I suspect there are a few more differences between how board.next() is implemented in python vs TS. Need more time to investigate this.

Sample runs:

miner vs do_nothing

node --require ts-node/register interpreter.ts run 20 miner do_nothing
> agent miner wins: 20/20

node --require ts-node/register interpreter.ts step 20 miner do_nothing
> agent miner wins: 19/20

I have observed that miner performs consistently slightly worse in step interpreter compared to python interpreter. I will investigate this further.

./main.py (bot) vs miner

node --require ts-node/register interpreter.ts run 20 ./main.py miner
> agent main wins: 00/20

node --require ts-node/register interpreter.ts step 20 bot miner
> agent bot wins: 00/20

I have not testing training with this new interpreter but this change is already very big and it is working fine on its own, so I think we can merge this first.

@bovard please kindly help review.

[bovard]

LGTM