Remove Prio3FixedPoint64BitBoundedL2VecSum

[divergentdave]

The instantiation of the fixed point vector VDAF with 64-bit fixed point numbers is of limited use, because it can only handle one- two-, or three-dimensional vectors.

failed to construct Prio3FixedPoint64BitBoundedL2VecSum VDAF

Caused by:
    0: flp error: value error: number of entries (4) not compatible with field size
    1: value error: number of entries (4) not compatible with field size

This is because the vector coordinates each get encoded in 63 bits, so the square of the vector's L2-norm gets encoded in 126 bits. Adding four squared coordinates together would bring you to $\approx 2^{128}$, which would exceed the prime modulus of Field128.

Given that the intended use case of this VDAF is federated learning, where low dimensionality and 64-bit fixed point precision are uncommon, I think we should remove this VDAF, leaving just the 16-bit and 32-bit versions. If need be, we can always add another specialization at, say, 48 bits down the line.

cc @ooovi @MxmUrw

[MxmUrw]

Yes, fair enough. I think we should do so after #1440, right?