Unified the paginators for pylons and YUI.
- YUI based paginators now behave more like the ones generated with pylons
- introduced new custom pylons paginator for customizations needed to unify both

invalidation: don't create CacheInvalidation records on startup

Creating the records early gave an advantage before lightweight was introduced.
With lightweight it is no longer necessary.

The records will be created on demand anyway and there is no reason to create and
maintain them before they are used.

invalidation: merge .invalidate and .set_valid as .test_and_set_valid

This simplifies the internal API and avoids leaking CacheInvalidation instances
and cache keys.

removed assertion about ancestor revision.
log this as a warning, and let it proceed.
ref #829

Removed shortlog aka lightweight changelog.
Current version of changelog is fast and condensed.
There's no sense to keep changelog duplicity.

Use changelog controller for displaying history of files.
step 1 for removing obsolete shortlog

api: don't report invalidated cache_keys after invalidating a repo

The keys are an implementation detail and is never relevant outside RhodeCode
internals and should not be leaked.

compare: move get_changesets to compare controller where it is used - 2nd half that was lost in 6c79bfcd3b54

That is a better location than the pull request model. It should probably be
split up and moved to the backend specific Repository classes.

moved around some code in changeset controllers to properly log which function was decorated.
Before it was always index, and permission were checked twice giving unnneded overhead

auth decorators are not used anymore on __before__
- this will allow to create a whitelist of API enabled controller functions
- moved around functions with _ to beginning of the files to make it cleaner

changelog pagination with branch filtering now uses
common logic that non branch filtered version.
- introduced CollectionGenerator object as a helper for
achieving abstraction layer on top of filtered results

get_changeset uses now mercurial revrange to filter out branches.
it's around 20% faster this way

pullrequests: fix reversed diff

b2575bdb847c missed the 'we swap org/other ref' part.

implemented usergroup permissions checks.
- each user who has now explicit or inherited permissions can create usersgrous
- user is automatically an admin of that usergroup, and can manage it

Iteration on default permissions
- added user groups default
- organized global permission form, and made it common for
users, and user groups
- form improvements and intructions

fixed default permissions population during upgrades
- it often happen that introducing new permission
caused default permission to reset it's state to installation
default.
new version makes sure that only missing permissions are
created while leaving old defaults

#818: Bookmarks Do Not Display on Changeset View
- made the consistent order of displaying tags iconss

#818: Bookmarks Do Not Display on Changeset View
- made the consistent order of displaying tags iconss

pullrequests: optimize locating changeset to compare

We only need the max revision - that is much faster when a branch is specified.

pullrequests: use hashes instead of changecontexts

The changecontexts was converted to hex strings when stringified, but better be
explicit.

pullrequests: handle the case where no matching revisions are found for either end

Use -1 as default revision - that is the null revision.

pullrequests: simplify handling of null revisions

Undo unnecessary and incorrect code introduced in bd5420ea396b.

compare: move get_changesets to compare controller where it is used

That is a better location than the pull request model. It should probably be
split up and moved to the backend specific Repository classes.

pullrequests: fix changesets ordering being reversed when creating new pull requests

41b4edf77b5b tried to make the displayed order of changeset consistent: The
topmost is always the latest.

That did however also reverse the ordering of the changesets sent back in the
post when used in the pull request creation form. Displaying the pull request
later on would reverse it again and thus show it in the 'wrong' order.

We now undo that reversing when creating the pull requests, and the stored data
will thus be the same as before.

diffs: drop diffs.differ

This function did not really add any value, it was yet another layer that was
hiding the bug that we use ref[1] without ref[0], and it had this strange
undefined behaviour for diffing across repos.

Inlining the call to .get_diff do not make the code more complex but makes it
more obvious what is going on so it can be cleaned up later.

consistent handling of grant/revoke of permissions widgets
- use 1 method for users and for users groups
- use common JS for ajax revoke
- cleanup code

- Manage User’s Groups: create, delete, rename, add/remove users inside.
by user group admin.
In this case, a user's group can be owned by several people thru an owner user's group.

Some refactoring of naming, permission handling logic.
- remove some code duplicity as well as inconsistent naming

shortlog: show all refs (bookmarks,tags,branches) in one dedicated column
- rename confusing name bookbook to booktag (as others similar names)

repo rescann should detach forks of zombie repos,
this way we ensure we do proper cleanup of such repos.

added __eq__ operation on vcs Repository objects
- eq compare if objects are same class and have the same path
- fixes some issues introduced after @LazyProperty was removed from scm_instance

invalidation: simplify get_prefix

str.split will always have a [0], and cache_key will always contain
repo_name=cache_args and the split will thus always return two elements