Cardano Updates

* Workaround for memory annotation problem

* Update golden files

* Remove Plain from ExMemory.hs

* stylish-haskell problem during merge

* stylish-haskell problem during merge

* Change * to Type

* Remove redundant import

This gives the ledger an interface that looks like a bag of named
integer parameters.

The implementation is ugly. I wanted this to work today, so I went with
something cheap and unpleasant: it would be nice to have something more
principled, but the current version does work.

We may want to revisit any number of these things. We can get rid of
some ugliness once we also use integetrs internally. But also stuff
like: the set of keys is derived from our current JSONification, which
isn't ideal, it should be highly stable. However, apparently it's okay
if we change the keys a bit before the testnet, so we can come back to
this.

Finally, I added some actual tests that the example scripts work, which
is helpful!

* SCP-2133: use exceptions in the CEK machine

This refactors the CEK machine to use exceptions instead of `ExceptT`.

This results in a substantial (30-50%) speedup on the validation
benchmarks. (Will post benchmark results in a comment.)

The cost is some evil in `runCekM` which we should inspect *very*
carefully. We also need some more type applications since we no longer
have the nice `MonadError` functional depenency that fixes the error
type.

So far I *haven't* pushed this down into the constant application
machinery. The main reason for this is that we have some awkwardness at
the interface of `applyTypeSchemed` where we need to change the `cause`
part of the error from a value into a term, and I don't want to have to
catch and rethrow an exception there if we can avoid it. It's also
convenient for other users of the CAM to have normal errors there. So
I'm not sure if we *should* change that.

* [Evaluation] [Refactoring] Worker-wrapper-transformed the CEK machine

* Comments

Co-authored-by: effectfully <[email protected]>

This gives the ledger an interface that looks like a bag of named
integer parameters.

The implementation is ugly. I wanted this to work today, so I went with
something cheap and unpleasant: it would be nice to have something more
principled, but the current version does work.

We may want to revisit any number of these things. We can get rid of
some ugliness once we also use integetrs internally. But also stuff
like: the set of keys is derived from our current JSONification, which
isn't ideal, it should be highly stable. However, apparently it's okay
if we change the keys a bit before the testnet, so we can come back to
this.

Finally, I added some actual tests that the example scripts work, which
is helpful!

Build the deployment machines as part of the CI

Previously all of the deployment processing was completely outside of the CI. While some nixos vm tests have already been added that at least verify the NixOS modules used in the configurations that we deploy, the actual machine derivations were never built. This PR adds deployment/morph/default.nix which builds all machines that are actually deployed.

Evaluate deployments to {} (empty attribute set) for darwin since our
release.nix doesn't filter this out and hydra fails to evaluate this on
darin (cannot build nixos machines on darwin of course).

Instead of building the whole attrset that `pkgs.nixos` returns just
pick the `toplevel` derivation. This way we only get a single derivation
instead of an attribute set which was causing problems when collecting
derivations for hydra.

Inject tfinfo so it is passed to all modules that get imported.
This is neccesary so that all machines can access terraform
provided by tfinfo.