Check consistency of mapping from path to rid and vice versa, also check
catalog metadata for its contents.

- Split catalog module into surgery and checkup.
- Use unhealthy instead of abberation
- Use healthcheck instead of checkup
- shorten zopectl command name

The object that was used to catalog these entries is gone and cannot be found
at path any longer.

Only commit in surgery and only if the post-surgery health check is successful.
Also add '--dry-run` argument to prevent committing for a testrun.

The UID index contains one UID per plone object. The UID for each object is
unique, thus we know that for each object in the catalog there will also be one
entry in the UID index. Thus we can use this index as a sample to check the
catalog index health.

Default dryrun to true for checkup. Checkup must never cause persistent changes
and thus never commit.

Remove manually from indices as unindex_object aborts too
early if the internal data is inconsistent.

We mostly check the catalogs internal data strucuture and not the indices. I
think not mentioning indices in the method name is better.

This hopefully helps a bit to avoid confusion between the uuid_index, an index
in the catalog and the catalogs internal uid attribute which contains the
path->rid mapping.

Consistently provide these methods as helper methods on the base test class.

Check rids in index and unindex against each other and agains the catalogs
uid mapping (path->rid).
This seems to change some of our assumptions regarding how we should update
catalog index length.

Creates an unhealthy catalog the same way as happens in production. Should then
be fixable by surgery.

Also document in what way the catalog is unhealthy and provide entry points for
further analysis while developing fixes.

Some indices seem to be ablte to maintain their internal integrity thus we can
use the official `unindex_object` method to remove a rid from these indices.

Allow to pass a `Length` that has to be decreased once an index row is empty.
Slightly refactors coupling of `index._length` to the index which is most
reliably used to determine the correct length as this varies heavily between
different index implementations.

Re-use existing surgery for extra rid, that is not in the UUID index, but in
rid-->path mapping and catalog metadata.

Some fixes reindex objects, to guarantee an up-to date catalog before
post-surgery healthcheck we process the indexing queue.

Correct special case handling for objects that were moved into one of their
parent objects and thus can still be found via acquisition when their former
path is used as their id was not changed.
We assumed that the redirector is involved, this is not correct. Thus drop
redirector check and instead make sure that the acquisition chain of the wrapped
and unwrapped object is different.

Explicitly doom transaction in healthcheck command, factor out healthcheck
helper function that can be called from surgery.

Improves readability by avoiding to overload the name surgery.

Also improve testing and add another assertion here and there.