Bug in pruning map over ZZ with torsion

Nobody has claimed this yet.

Still haven't looked into the root cause, but I think this is a bug:
```m2
N = prune coker matrix{{2, -1}, {-1, 2}, {-1, -1}}
-- cokernel | 3 |
-- | 0 |
f = inverse N.cache.pruningMap
-- | 1 2 0 |
-- | 1 1 1 |
g = map(target f, source f, matrix f)
-- | 1 -1 0 |
-- | 1 1 1 |
f == g -- false
```
The matrices are not the same over the integers, but as morphisms of Z-modules they are identical. In particular, I think one would expect `f == map(target f, source f, matrix f)` to always be true.

The closest I got to directly defining such matrices is something like this, which returns a non-sense error:
```m2
i1 : debug Core;

i2 : R = ZZ/3;

i3 : m = map(R, raw matrix{{2}})

o3 = | 2 |

1 1
o3 : Matrix R <--- R

i4 : n = matrix(R, {{2}})

o4 = | -1 |

1 1
o4 : Matrix R <--- R

i5 : m == n
stdio:5:3:(3): error: expected same ring

i6 : ring m === ring n

o6 = true
```

Start by reproducing the pruning and torsion examples in the issue, then inspect the behavior of `inverse N.cache.pruningMap` and `map(target f, source f, matrix f)`. Done means the reported morphism comparison and the related matrix comparison behave consistently, with the nonsensical ring error explained or corrected.

Written by the indexing model from the issue text.