Combine lambdas in OCaml output

[elegios]

This PR uses syntactic sugar in OCaml to write, e.g., let f = fun a -> fun b -> ... as let f a b = ..., and similarly for merging anonymous functions. This has some benefits:

• Reduces whitespace, before we would indent one level per let and fun, now just once for let.
• Shorter code, might be easier to read.
• After switching from dune to ocamlfind (Use ocamlfind instead of dune in compile.mc #873) I eventually found that backtraces in gdb had every function named as just fun, i.e., it only saw anonymous functions. I have no idea how the two tools invoke ocamlopt differently to get that effect, but merging arguments like done in this PR recovers a useful backtrace.

[elegios]

We concluded during the meeting we wanted to see how this might impact performance. Here's the bootstrapping process before:

 2) [163.162s] [build] src: ./boot eval main/mi-lite.mc -- 0 main/mi-lite.mc ../build/src/mi-lite        
 1) [49.822s] [build] src: ./mi-lite 1 main/mi.mc ../build/src/mi1                                       
 0) [50.277s] [build] src: ./mi1 compile main/mi.mc --output ../build/src/mi 

and after:

 2) [153.074s] [build] src: ./boot eval main/mi-lite.mc -- 0 main/mi-lite.mc ../build/src/mi-lite        
 1) [31.709s] [build] src: ./mi-lite 1 main/mi.mc ../build/src/mi1                                       
 0) [31.805s] [build] src: ./mi1 compile main/mi.mc --output ../build/src/mi

I've done a few runs, and the first step seems to be roughly equal time taken in the two versions, while step 2 and 3 seem to consistently go from ~50s to ~30s, which is... weird, to put it mildly. The intermediate .ml file that's passed to ocamlopt goes from 33MB to 25MB.