0.1.0

Make a lot of breaking changes.

I'm hoping no one was actually using this before this commit, and if you
were, I apologise in advance.

I really should have committed more often. Instead, I give you this
atrocity. I'll try to document as much of the changes as I can remember:

* Use the `windows-sys` crate instead of dumb `#[link(name = ...)]` and
  `extern "C"` definitions.
* Ensure that no sneaky `bcrypt.dll` sitting alongside the binary gets
  loaded instead of the system one by setting proper linker flags.
* Unify the set-up binaries into one. Unfortunately we have to keep the
  task runner separate because we don't want it to pop up a console
  every time it runs.
* Release SHA256 hashes of the ZIP archives.
* Build releases for AARCH64 and i686 targets.
* Change the location of logs and configs to `$HEARTBEAT_HOME`,
  (`~/.heartbeat` by default), to be consistent with other clients and
  the server implementation. We aren't adhering to standards here,
  however, this makes imploding ourselves easier because when you don't
  want to use heartbeat or anything related to it anymore, all you have
  to do is nuke this one folder. This gives me one less thing to worry
  about when I get around to building an installer for this.
* Fix the task getting run more frequently than it should because
  triggers step on each other. Since it's configured to run only while
  logged in, and the check for workstation lock happens on each run, we
  can just configure it to have a single trigger on a schedule, starting
  ASAP.
* Change the top level config key from `heartbeat` to `client`. This
  might come in handy if I ever want to unify the server and client
  config (though why you would want to run both on the same device beats
  me)
* Actually check that the task runner executable is present alongside
  the `heartbeat-client` executable during configuration.
* Avoid shelling out to wmic for user information. This did, however,
  lead me to write a lot more unsafe code, but it's sound enough.
* Fix the task XML template to placate newer Windows 11 builds, which
  die if tags aren't in the right order, for some reason. Bonkers.

I think that about sums it all up. And, regrettably, I cannot promise
that things like this will not happen again.

0.1.0a1

remove structopt due to dependency on atty

0.1.0a0

prepare for release