Perfection is a goal in itself - use arbitrary precision numbers for time tracking

[jrkatz]

After running for a long while, we should expect small timing errors to accumulate (float-truncation issues), so while no individual beat was off by much, the track as a whole is not where we'd expect it timing-wise.

Contrary to the title, we need not use arbitrary precision if we can find a clever way to minimize or balance the accumulation of floating point errors. So that's an option. Or we can fit in a library like https://github.com/MikeMcl/big.js/, or roll our own arbitrary precision numbers.