M2: Apparently it's not possible to recover the original spin, so I guess you just end up with a random spin.

Stern-Gerlach Interferometer with Realistic Magnetic Field:

Abstract: In this report, we investigate the spin dynamics of a neutron beam in a Stern-Gerlach experiment. In contrast to the simple constant gradient magnetic field assumed in most literatures which violates Maxwell Equations, we work with a model of magnetic field which satises Maxwell Equations. The spin dynamics is investigated by solving the Schrodinger equation using Fast Fourier Transform method. The spin coherence of the neutron beam is found to exhibit the humpty-dumpty behaviour[1, 2, 3] even though there is no fluctuation to the magnetic field nor any environmental noise. The main cause of the spin decoherence is identified as the inhomogeneity of the magnetic field in the x and y components. In addition, the nonlinear terms in the z component of the magnetic field, zn, n > 1 also contribute to the loss of the spin coherence. Although only magnetic field model considered is very specific, the cause of the loss of spin coherence identified using this model is a common features of any realistic magnetic to be used in a SGA experiment. Therefore the humpty-dumpty nature of the spin coherence exist even without any fluctuation to the magnetic field and is inherent to the SGA experiment itself.

And see also this earlier paper, Is spin coherence like Humpty-Dumpty? I. Simplified treatment (free full text), which gives a more understandable qualitative argument:

Why should it be so difficult to maintain spin coherence? The essential features of the following qualitative argument are due to Heisenberg.

I won't try to quote the actual argument, which is on page 3 of the PDF, but it seems that the main problem isn't entanglement between the particle and the apparatus.