Build hardware to support required low latency data

[gnperdue]

Once we identify required low latency data streams (see #7), we must build or procure the hardware to support getting the data to the board.

Note - there could be more than one iteration of this (training based on historical data may not have the same dependencies as training with low-latency data), so we should be careful to make the solution extensible.