During startup open a connection to every configured database, to ensure all of them can indeed be accessed. This avoids surprises during runtime when traversal to some database mountpoint could fail as the underlying storage cannot be opened at all.

Explicitly close all databases on shutdown, which ensures `Data.fs.index` gets written to the file system.

Move Zope2 tests at the end, they tend to leave the process in an unclean state and cause reference couunting / garbage collection timing issues in AccessControl tests

Removed the old help system, in favor of the current Sphinx documentation hosted at http://docs.zope.org/zope2/. For backwards compatibility the `registerHelp` and `registerHelpTitle` methods are still available on the ProductContext used during the `initialize` function.

Removed various persistent product related code and options. The `enable-product-installation` `zope.conf` setting is now a no-op.

Removed the `enable-ms-author-via` directive which was only required for very old web folder implementations from before 2007.

Remove various rarely used options from the example zope.conf.in - they are still present in the zopeschema.xml

Changed zope.conf default settings for `zserver-threads` to `2` and `python-check-interval` to `1000`.

Simplified instance skeleton, removing old `Extensions`, `import`, `lib/python` and `Products` from the default. You can continue to manually add these back.

Removed the rarely used support for the `++skin++` traverser and added instructions on how to get it back for a specific application

Remove duplicated changelog entries for changes also made in 2.13.x - I want to see what's new on trunk

Previously an addAfterCommitHook was used, but this is not run on transaction
aborts.  Now a Synchronizer is used which unconditionally closes the request
after a transaction is finished.

Not closing the request seems to have caused the database connections to be
kept open and these nasty log messages:

    CRITICAL ZODB.DB DB.open() has 42 open connections with a pool_size of 7

42 being the wrong answer in this case.

Make Yvo happy and keep testing our factored out distributions, so CMF doesn't have to grow a KGS of its own