Add Feynman path integral backend

[Renmusxd]

The existing backend is "Schrodinger"-style, which means it stores the entire wavefunction, runtime is O(2^N M) and space is O(2^N) where N is the number of qubits and M the number of gates. A Feynman path integral backend would sum up all the possible paths from the input to the output through the circuit, making the runtime exponential in the circuit depth, but the memory footprint constant. This may be difficult to implement because of measurements taken midway through the circuit, but would be very helpful for certain circuit types. Furthermore there are likely mixed S-F algorithms which can optimize space/runtime tradeoffs.

[Renmusxd]

Some work done on this, a basic version with no memory has been implemented but that's very computationally expensive for most graphs. Memoization helps this significantly and a simple version was added in 24cd7f1 but it's not very smart.

[Renmusxd]

Out of date as of rewrite.