and Cap (to fix)

  to detect that we have the most precise solution
* Apply clean_type more agressively but on smaller types, for each of the partial solutions
  rather than on the (big) intersection of all the solutions.
* Small optimisation in substitutefree: do not do anything if the type has no variables

- fix a bug in Positive.decompose that was introducing components that were not
in the original type
- rewrite Positive.decompose to flatten long chains of `Cup [`Cup[`Cup ...]]
- rewrite apply_raw to perform the correct dove-tail exploration of the branches
- rewrite Positive.clean_type to perform substitution of {co,contra}-variant only
variables and internal variable beautification at the same time
- For the moment, only call Positive.clean_type at the very end, before returning the result type. Need to investigate if intermediate results would benefit from having less variables.

Do not call Arrow.apply since the types bound to the variable Gamma will be the result of the application.

- fix typos in comments
- indent code
- move the Positive module further down types.ml, so that there is no
need for forward references when we implement substitutions over record types.

 * make sure (decompose t) builds a cyclic 'v' object if t is a recursive type.
 * add a memoization environment to substitute_aux to detect and rebuild such cycles in the new 'v' object.
 * use solve everywhere to build a cycle type (this basically implements Courcelle's algorithm at the level of types.

* More agressive simplifications of unused variables, and variable beautification of remaining variable (i.e.
rename _fresh_xxx, _fresh_yyy into $A, $B, ...

	add operator "=" for integers;
	add constant values "`true" and "`false" as possible patterns

Conflicts:
	types/types.ml
	types/types.mli