+ fix a bug in "compare"

+ rename Debug.Opt to Debug.Args to avoid conflicts

+ create AUTHORS file
+ fix the linking exception in LICENSE
+ update the "About" in IDE
+ remove the trailing whitespace
+ inflate my scores at Ohloh

Change session serialization so that element ordering no longer depends on hash values but on actual values.

and a bit of cleaning at the same time

But:
 - problem with BuiltIn, Bool, Tuple, Unit and theory
 - work only with "why" and "whyml" files format

(in case of a prover upgrade)

  - the symbols that appear in the metas are identified in the xml by
    their position in the task:
    - in which declaration
    - in which definition (if that apply otherwise -1)
    - in which constructor(or case in inductive predicate) (if that apply otherwise -1)
    - in which field (if that apply otherwise -1)

  - the md5sum of the prefix of the task that end with the declaration is used to know if the
    symbol have been changed, and if it is obsolete.

  - currently metas that contains obsolete symbol are removed.

and Debug.register_stop_flag into register_flag. While information
flags (selectable by --debug-all) are more common, it is safer to
treat a generic debug flag as behaviour-changing by default.

comportement was not very clear if an exception was raised.

 (metas, debug flags, transformations, formats) except for label.

This description is used in --list-*. The description can use any of
the formatting markup of Format "@ " "@[",...

Transformations can also specify from which metas and labels they
depend, and add informations about how they are interpreted.

TODO:
  - complete and correct the documentation
  - when a transformation use Trans.on_meta, it should be possible to
    add an interpretation of the metas in the documentation.
  - recover a summary version of --list-* ?
  - be able to export in latex?

  session pairing doesn't compute anymore the shape of the goal, it is
 done before. It was able to compute the shape only when the checksum
 of the task was different, but computing the checksum of the task is
 way more time consuming than computing the shape of the goal (and
 include it).

 So this commit simplify greatly the function and theoretically
 augment just a little the time spent. Experimentaly it's the inverse
 on max_matrix. Until "update_session: done" with or without modifying
 the checksums:

            before     |   after
without : 0.21-0.22 s  | 0.16-0.17 s
with    : 0.23-0.26 s  | 0.18-0.20 s

the index in the hastbl of proofs on parent goal must be
changed too

The goal is to keep track of unexpected termination, especially when the
output of the prover is empty.