Extended order_by and bugfixes

[fabiansinz]

• relation.fetch.order_by allows for individual sorting arguments now. I removed the kwarg descending. The new syntax is relation.fetch.order_by('name DESC', 'trial', 'stimulus DESC'). If the sorting argument is omitted, ASC is set by default.
• I removed the relation.fetch.apply function. It is not needed at the moment and simplified the syntax substantially. Additionally, it might cause confusion, since something like relation.fetch.apply(myfunc).order_by('stimulus')['name'] suggests that the function is applied in the beginning, but it would have been applied at the end.
• I renamed FetchQuery to Fetch and Fetch1Query to Fetch1
• I renamed limit_by to limit_to. Dimitri actually prefers limit, but I think limit_to is more consistent with order_by.
• Wrote quite a few more tests.

[dimitri-yatsenko]

note that order_by can operate on expressions, not just attribute names. For example, if I wish to sort so that name='Fabian' comes before all others I would need to say, according to the proposed syntax:

relation.fetch.order_by('name="Fabian"=DESC', 'trial')

This does not seem like good syntax. Plus the word DESC is hardly PEP8-friendly.

How about tuples ('expression', True) for descending.

relation.fetch.order_by(('name', True), 'trial', ('stimulus', True))

[dimitri-yatsenko]

What's the reason for copy_first?

[eywalker]

We thought that it would make more sense for each operation on the fetch object to return a new fetch object rather than modifying the object in place. Since fetch object in itself is very cheap, this shouldn't be a real over head. copy_first is a decorator that modifies the decorated method to work on and modify the copy of the object, supporting this idea.

[dimitri-yatsenko]

why not use copy.copy?

[fabiansinz]

copy.copy does not copy the dict behavior. The function deepcopy would copy _relation as well which seemed like an overkill. That's why I introduced a copy constructor and used a decorator to tell the function, that I'd like self copied. Having the elements of behavior as object attributes wouldn't solve the problem, since some of the values of behavior are lists.

[dimitri-yatsenko]

But why do we need to copy the object in the first place?

[dimitri-yatsenko]

ok, i have looked into other modules and, yes, things like these usually copy the object. I am merging this PR.