plutus-ir: allow lets to be strict or non-strict

[michaelpj]

This implements a feature I discussed before on the mailing list: non-strict (possibly should be CBN) let-bindings in PIR.

This has a couple of advantages:

• We can simplify the plutus-tx code significantly, removing some restrictions in the process.
  • This removes most of the annoying overhead introduced by plutus-tx: compile let-bindings of non-values lazily #1019
• We can simplify the recursion-thunking code, at the cost of being slightly less clever.
  • Previously we introduced an "adaptor" binding so that the body of the let would see a non-thunked binding. Using non-strict let-bindings means that the body will now see the thunked binding, but this doesn't matter that much in practice.

There is one con at the moment:

• We generate the unit type and values at each use site in PIR, since we don't have definitions support there. I should fix this, but I don't know how much this will actually save.

I also fixed the value check in PIR. In particular, it thought that variables were values, which is definitely wrong. As a result we now compile a lot more cases non-strictly, in particular a lot of trivial cases on booleans. But again, I think this doesn't matter so much and it's much more correct.

We still have logic for compiling cases non-strictly in plutus-tx. I want to add case expressions to PIR to handle this, but I'll do that in a separate PR.

[effectfully]

Great!

[effectfully]

This is called "simultaneous substitution", I believe.