Add additional regression tests for --replace-function-body

[hannes-steffenhagen-diffblue]

As per @thk123 s comments on #1889 there are some additional edge cases which we should be testing for, for instance pointer-to-pointer-to-const.

[martin-cs]

Particular horrors are the difference between 'const int *' 'int * const' 'const int * const' and structures (and unions?) with some const fields and some non-const ones.

[hannes-steffenhagen-diffblue]

@martin-cs What should the behaviour in that case be? Just refuse to assign, or assign to all non-const fields manually?

EDIT referring to structs with a mix of constant and non-constant members

[martin-cs]

@hannes-steffenhagen-diffblue I'm not sure I have one fixed answer in mind; just as long as there is a consistent, correct and justifiable answer. Throwing an error, assigning all non-const fields (possibly with a warning), assigning everything (with a warning), these all seem possible.