Need tests of PREPARE

[zachmu]

#509 demonstrates that we have bugs that are only triggered on statement prepare, which we have no explicit tests of. Essentially anywhere in the analyzer that does type assertions on literal values needs to consider the possibility of a bindvar being there instead. There are probably lots of bugs here.

To fix this, we need explicit tests of prepare, with bind vars in every place they can appear in a query. Probably this is easiest to accomplish by creating a new harness variant.

[zachmu]

We've made enough progress here (prepared forms of most engine tests) that we can close this out.