Merge lp://qastaging/~thisfred/desktopcouch/remove-recordtype-from-dict into lp://qastaging/desktopcouch

Fixes #674487 where the record_type ended up in the dictionary of the record, where almost all client code had to filter it out again.
Fixes #669133 because the record_type version was never saved to CouchDB.
Fixes #575772 where the couchdb attachments API was defined on the wrong class.
Fixes #675787 where MergeableList.pop() could not be called without an index.

Note that the frustration is mostly aimed at myself, since this shows that I suck at reviewing and writing code, since I did at least one of those for all of the code that was buggy.

I like this branch. A couple comments:

- since you don't need current_value anymore here:

- for index, current_value in enumerate(self):
+ for index, _ in enumerate(self):

you might as well write:

     for index in range(len(self)):

Also, what's with the double blank lines before or after functions and methods? You only need double between classes. Don't get overboard with that automatic pep8 thing. ;-)

> - since you don't need current_value anymore here:
>
> - for index, current_value in enumerate(self):
> + for index, _ in enumerate(self):
>
> you might as well write:
>
> for index in range(len(self)):

Yeah that's better

> Also, what's with the double blank lines before or after functions and
> methods? You only need double between classes. Don't get overboard with that
> automatic pep8 thing. ;-)

Actually I like having two lines between top level definitions no matter what they are (so functions or Classes, *not* methods.) It makes me have to think less, and improves readability. Also a top level function and a class definition are rather similar things IMO.