Compiler: Rework inlining

[hhugo]

Should fix #357

@vouillon, I've rebased an old branch of yours (aggressive-inlining) and changed the logic a bit so that

• it can replace the previous "simple" optimization (inlining closure of "size" one).
• it only duplicates other small closures if they call one of their arguments in tail position (this is an attempt at addressing Better inlining ? #357)

[hhugo]

@pmwhite, @TyOverby, you might want to try this PR as it could affect performance and size of artifacts.