Signed-off-by: François Revol <revol@free.fr>

Don't force objects to have an id, especially since that id was
incorrect. "plap/plop" would have the same id as "plop", i.e. "plop".
We don't rely on the id for handling Collections anymore.

Change to use more unique property names.

Change the display, instead of using '*', we use '~', and there is no
"Collection" header anymore.
Updated formatters could use that way of showing the object is also a
collection too.

refs #774

There isn't enough code to justify multiple files.
Also includes pep8 fixes.

* Create a get_collection method similar to get_* methods for objects.
* Fix title initialization of a collection
* Remove the hack were both id and title were allowed when CDing.
  That hack only worked with the canalplus module, and failed with
  others like redmine (but they still showed in suggestions).
  Moreover, the canalplus module now has friendlier IDs
  so this is not really needed anymore.
* Allow backends to tell if a path is valid or not.
  For instance, it now allows to cd in a Redmine project with no issues
  in it. It also won't display "404" for invalid project IDs.
  By default, we still use the unreliable method of checking there is at
  least one result in iter_resources().
* Fix cd completion to work with unicode strings (all strings after an
  unicode string were ignored!)
* Do not suggest '..' when completing cd in the root

refs #774

* Remove callbacks in Collection object
  Make Collection a "dumb" object (and also a base object,
  though it isn't very useful for now)
* Rename Path to WorkingPath, because it is more about managing state
  than being a single path.
* Rewrite almost all WorkingPath, because the code was overly
  complicated for no reason (I tried some special cases and it turned
  out that fromstring didn't handle them, and that the
  quote-escape-unquote was just unecessary). I also rewrote it to be
  more pythonic (no more lambdas and maps) and added tests.
* Require the full split path when creating a Collection. Because, come to
  think of it, an object needs an unique identifier; in the case of
  Collections, it is the full path, not only its last part.
  I might even replace the id by the full split path in the future.
* There is now only one way to get items of a Collection: calling
  iter_resources().
* Rewrite flatten_resources to iter_resources_flat(), which just calls
  iter_resources() recursively.
* Rewrite the collection part of the canalplus module. There is no more
  callback or a page calling the browser to check another page!
  The logic is only in iter_resources().
  The resulting code is not very pretty, but it should get better.
  As a bonus, avoid to reload the main XML file when we already have it
  open.
* change_path() now expects a split path and not a string.
* up/home special cases for "cd" are handled in the same place, and
  store the previous place properly (but are not yet exploitable by
  an user command).

This is a big commit but it would be hard to split it in *working*
commits.

If you read this entire commit message, I will buy you a beer.

refs #774
fixes #773

* Make the declaration of fct and it in the constructor Collection,
 instead of adding them from the outside
* Add a function to flatten a list containing collection (solves the
 radioob search crash)
* Better display of collections in the "ls" command (and display both id
 and title)
* The "cd" command goes to the root of the path (like the UNIX cd)
* Move the Video object of canalplus in a correct path
* Make Collection iterable
* Add comments to CapCollection
* Cache the result of fct in a Collection; it is only called once
* CollectionNotFound errors can be more explicit by providing a path
* Require utf-8 in collection paths
* Code cleanups