Implement observables API and autocorrelation time computation

[RadiatedMonkey]

Implements the new observables API and automatically computes autocorrelation times for observables.

An observable can be created by implementing the Observable trait for a struct. This struct can then be added to the system using the SystemBuilder::with\_observable function. The code makes use of a compile time HList to store the observables which means there is no runtime dynamic dispatch involved.

After the system has thermalised, the system will measure all observables on every sweep. This information is then used to compute the autocorrelation time for each observables using the FFT method. This method works as follows:

Collect all observables into a time series and then subtract the mean of the *entire* series from each entry to normalise it. We then double the list in size, filling the remainder with zeroes. After this we take the Fast Fourier Transform of this list and multiply the last by its conjugate. We then take the inverse FFT of this list to obtain the autocorrelation.

To get the (integrated) autocorrelation time we finally divide the whole list by the first element and compute

$$\\tau = \\frac{1}{2} + \\sum\_{i = 1}^N \\rho(i)$$

This autocorrelation time tells us how many sweeps to discard between measurements.

To determine N we use the Madras-Sokal window method. This method reduces the noise that is introduced by higher frequencies. For this we choose some W such that

$$W = c \\tau\_{\\text{int}} (W)$$

for c approximately 5. This W is found by starting with W = 1 and incrementing until the condition is satisfied. Once the autocorrelation times are established, measurements will only be taken every 2 autocorrelation times for each observable.

While the simulation is running, we then recompute the autocorrelation times periodically.

Closes #9
Closes #8