To reconstruct a sine wave you need two samples per period.
If we are being scrupulous, this is not entirely correct, as it depends on the signal phase.
For example, let's say we have a 1 GHz sample rate and two samples (see picture below): 0, 0 (sin(0)=0, sin(pi)=0)
It's obvious that you cannot distinguish whether it is a sine wave or DC.
And if we know that this is a sine wave, its amplitude still unknown.
So we have two samples per period and know that the signal is within the Nyquist bandwidth (Fs/2), but we are unable to reconstruct its waveform.
But if we shift the signal phase by 90 degrees, we get the samples +1, -1, which allows us to reconstruct the original sine waveform.
Therefore, it would be more accurate to say that
two points are not enough to reconstruct the shape of a sine wave, but it is enough to reconstruct the shape of a cosine wave.
Regarding to oscilloscope requirements, as a rule of thumb, the oscilloscope's sample rate should be at least 10-20 times higher than its analog bandwidth. This allows the anti-aliasing filter slope to remain within the Nyquist bandwidth and ensures good flatness within the oscilloscope's analog bandwidth.