Rasmey correction not working as expected using signal experiment

[andrea-pasquale]

The ramsey_signal experiment could give a wrong correction in frequency.
I think that the problem originates from the fact on the y axis we are using the amplitude $m = \sqrt{I^2+Q^2}$, and before running single shot classification we don't know whether $m_0 > m_1$. I don't know if there is short term solution for this.
We should have encountered the same problem in flipping_signal so perhaps we could take inspiration from that to find a solution:

qibocal/src/qibocal/protocols/flipping_signal.py

Lines 234 to 238 in d5d9191

	# TODO: this might be related to the resonator type
	if popt[3] > np.pi / 2 and popt[3] < 3 * np.pi / 2:
	signed_correction = translated_popt[2] / 2
	else:
	signed_correction = -translated_popt[2] / 2

To make the code more robust we should try to write tests that tackle all possible scenarios described above.