- build on nightly until rust 1.39 arrives
- switch to 6.1 API and create database directly
- drop support for Cluster and FDB releases prior to 6.1, as
the async changes are a breaking change anyway
- updated db.transact to retry failures in body, as in
#125
- made should_retry() on error objects public
- changed some sync functions that were using .wait() to async
(eg HCA allocate)

When fdb_future_destroy() is called on a future that hasn't resolved,
it causes the callback to fire. If it fires after the box the waker/task
was held in has already been deallocated, it can result in a crash.
Fix the issue by dropping the allocation in the callback instead.

This issue also exists in the stable foundationdb-rs release.

This reverts commit 371c34d.

While rust-toolchain remains in the repo, stable and beta tests
seem to run on nightly as well, so we should be able to leave the
tests there and they'll automatically start running on beta/stable
again when rust-toolchain is removed after the rust 1.39 release.

#135 (comment)