Merge `HostFixedTensor`, `MirFixedTensor`, and `RepFixedTensor`

[mortendahl]

All three are structurally identical to

pub struct RepFixedTensor<RepRingT> {
    pub tensor: RepRingT,
    pub fractional_precision: u32,
    pub integral_precision: u32,
}

so maybe we should simply have a single SpecificFixedTensor instead. Naming wise, perhaps this should be FixedTensor, and the current FixedTensor becomes AbstractFixedTensor to highlight that it's an abstraction.

Note that a team discussion needs to happen before implementing this, once the implications of the changes are fully understood.