Rewritting Bags from scratch using a namedtuple approach, along with other (less major) updates.

New bag implementation improves a lot how bonobo works, even if this is
highly backward incompatible (sorry, that's needed, and better sooner
than later).

* New implementation uses the same approach as python's namedtuple,
  by dynamically creating the python type's code. This has drawbacks, as
  it feels like not the right way, but also a lot of benefits that
  cannot be achieved using a regular approach, especially the
  constructor parameter order, hardcoded.
* Memory usage is now much more efficient. The "keys" memory space will
  be used only once per "io type", being spent in the underlying type
  definition instead of in the actual instances.
* Transformations now needs to use tuples as output, which will be bound
  to its "output type". The output type can be infered from the tuple
  length, or explicitely set by the user using either
  `context.set_output_type(...)` or `context.set_output_fields(...)` (to
  build a bag type from a list of field names).

Jupyter/Graphviz integration is more tight, allowing to easily display
graphs in a notebook, or displaying the live transformation status in an
html table instead of a simple <div>.

For now, context processors were hacked to stay working as before but
the current API is not satisfactory, and should be replaced. This new
big change being unreasonable without some time to work on it properly,
it is postponed for next versions (0.7, 0.8, ...). Maybe the best idea
is to have some kind of "local services", that would use the same
dependency injection mechanism as the execution-wide services.

Services are now passed by keywoerd arguments only, to avoid confusion
with data-arguments.

tests/config/test_services.py

@@ -3,9 +3,9 @@ import time
 
 import pytest
 
-from bonobo.util import get_name
-from bonobo.config import Configurable, Container, Exclusive, Service, requires
+from bonobo.config import Configurable, Container, Exclusive, Service, use
 from bonobo.config.services import validate_service_name, create_container
+from bonobo.util import get_name
 
 
 class PrinterInterface():
@@ -30,7 +30,7 @@ SERVICES = Container(
 class MyServiceDependantConfigurable(Configurable):
     printer = Service(PrinterInterface, )
 
-    def __call__(self, printer: PrinterInterface, *args):
+    def __call__(self, *args, printer: PrinterInterface):
         return printer.print(*args)
 
 
@@ -51,15 +51,15 @@ def test_service_name_validator():
 def test_service_dependency():
     o = MyServiceDependantConfigurable(printer='printer0')
 
-    assert o(SERVICES.get('printer0'), 'foo', 'bar') == '0;foo;bar'
-    assert o(SERVICES.get('printer1'), 'bar', 'baz') == '1;bar;baz'
-    assert o(*SERVICES.args_for(o), 'foo', 'bar') == '0;foo;bar'
+    assert o('foo', 'bar', printer=SERVICES.get('printer0')) == '0;foo;bar'
+    assert o('bar', 'baz', printer=SERVICES.get('printer1')) == '1;bar;baz'
+    assert o('foo', 'bar', **SERVICES.kwargs_for(o)) == '0;foo;bar'
 
 
 def test_service_dependency_unavailable():
     o = MyServiceDependantConfigurable(printer='printer2')
     with pytest.raises(KeyError):
-        SERVICES.args_for(o)
+        SERVICES.kwargs_for(o)
 
 
 class VCR:
@@ -100,13 +100,13 @@ def test_requires():
 
     services = Container(output=vcr.append)
 
-    @requires('output')
+    @use('output')
     def append(out, x):
         out(x)
 
-    svcargs = services.args_for(append)
+    svcargs = services.kwargs_for(append)
     assert len(svcargs) == 1
-    assert svcargs[0] == vcr.append
+    assert svcargs['output'] == vcr.append
 
 
 @pytest.mark.parametrize('services', [None, {}])