bugfix: copy-instance was not robust enough

[macrologist]

This partially addresses one source of "nondeterminism" in quilc compilation (see #676) that was due to a violation of referential transparency.

Specifically, non destructive compilation was not non-destructive enough: compiling the same parsed-program instance twice produced two different outputs. This was because copy-instance did not copy instances of measurement deeply, resulting in different outputs from compress-qubits, with downstream consequences.

This only addresses #676 in those cases where you are not parsing the same raw text-based source everytime you compile. Or equivalently, when you are not rebuilding your Program object from pyquil every time.

[macrologist]

(oh: I commented out the clifford tests b/c they've been failing for a while now and I wanted to sanity check my change, we might consider this PR, and any other wide-range-effect PRs to be blocked until the clifford modules and tests are loading, building, and passing again --- I think its just a quicklisp issue; waiting on an update.)

[stylewarning]

what a great find

[stylewarning]

re-enable clifford tests and we can merge

[macrologist]

turns out the cliffort tests are failing b/c the fucntion cl-permutation::iota-vector does not produce values of type cl-permutation::raw-perm. the former is being used to to produce permutation representations passed in to cl-permutation::%make-perm, throwing a type error. Sit tight.

[macrologist]

should be good