Previously they did not, leading to a reference leak of a tuple.

Fixes #216

Cooperate with InterfaceClass to ensure there is no performance penalty for this. Fixes #3

+-------------------------------------------------------------+----------------+------------------------------+------------------------------+
| Benchmark                                                   | bench_master38 | bench_issue3                 | bench_issue3_opt             |
+=============================================================+================+==============================+==============================+
| call interface (provides; deep)                             | 369 ns         | 454 ns: 1.23x slower (+23%)  | not significant              |
+-------------------------------------------------------------+----------------+------------------------------+------------------------------+
| call interface (provides; wide)                             | 373 ns         | 457 ns: 1.22x slower (+22%)  | 365 ns: 1.02x faster (-2%)   |
+-------------------------------------------------------------+----------------+------------------------------+------------------------------+
| call interface (no alternate, no conform, not provided)     | 671 ns         | 760 ns: 1.13x slower (+13%)  | 636 ns: 1.06x faster (-5%)   |
+-------------------------------------------------------------+----------------+------------------------------+------------------------------+
| call interface (alternate, no conform, not provided)        | 395 ns         | 494 ns: 1.25x slower (+25%)  | not significant              |
+-------------------------------------------------------------+----------------+------------------------------+------------------------------+
| call interface (no alternate, valid conform, not provided)  | 250 ns         | not significant              | 227 ns: 1.10x faster (-9%)   |
+-------------------------------------------------------------+----------------+------------------------------+------------------------------+
| call interface (alternate, invalid conform, not provided)   | 348 ns         | 424 ns: 1.22x slower (+22%)  | not significant              |
+-------------------------------------------------------------+----------------+------------------------------+------------------------------+

Only catch AttributeError instead of everything.

Fixes #200

Note that this does break a doctest in five.intid (it's expecting a TypeError but it now gets Acquisition's RuntimeError).

And fix the C handling of this case.

Several places needed to, essentially, call super.

This makes the rest of the attribute access fast again, but slows down
__module__.

+-------------------------------------------+------------+-------------------------------+
| Benchmark                                 | 38-master3 | 38-faster-descr               |
+===========================================+============+===============================+
| read __module__                           | 41.1 ns    | 123 ns: 2.99x slower (+199%)  |
+-------------------------------------------+------------+-------------------------------+
| read __name__                             | 41.3 ns    | 39.9 ns: 1.04x faster (-3%)   |
+-------------------------------------------+------------+-------------------------------+
| read __doc__                              | 41.8 ns    | 42.4 ns: 1.01x slower (+1%)   |
+-------------------------------------------+------------+-------------------------------+
| query adapter (no registrations)          | 3.85 ms    | 2.95 ms: 1.30x faster (-23%)  |
+-------------------------------------------+------------+-------------------------------+
| query adapter (all trivial registrations) | 4.59 ms    | 3.67 ms: 1.25x faster (-20%)  |
+-------------------------------------------+------------+-------------------------------+
| contains (empty dict)                     | 136 ns     | 54.8 ns: 2.48x faster (-60%)  |
+-------------------------------------------+------------+-------------------------------+
| contains (populated dict)                 | 137 ns     | 55.7 ns: 2.46x faster (-59%)  |
+-------------------------------------------+------------+-------------------------------+
| contains (populated list)                 | 40.2 us    | 2.86 us: 14.03x faster (-93%) |
+-------------------------------------------+------------+-------------------------------+

Not significant (1): read providedBy

This is pretty, but it slows down all attribute access to interfaces.
By up to 25%. I'm not sure that's acceptable for things like
Interface.providedBy.

+-------------------------------------------+------------+-------------------------------+
| Benchmark                                 | 38-master3 | 38-faster3                    |
+===========================================+============+===============================+
| read __module__                           | 41.1 ns    | 44.3 ns: 1.08x slower (+8%)   |
+-------------------------------------------+------------+-------------------------------+
| read __name__                             | 41.3 ns    | 51.6 ns: 1.25x slower (+25%)  |
+-------------------------------------------+------------+-------------------------------+
| read __doc__                              | 41.8 ns    | 53.3 ns: 1.28x slower (+28%)  |
+-------------------------------------------+------------+-------------------------------+
| read providedBy                           | 56.7 ns    | 71.6 ns: 1.26x slower (+26%)  |
+-------------------------------------------+------------+-------------------------------+
| query adapter (no registrations)          | 3.85 ms    | 2.95 ms: 1.31x faster (-23%)  |
+-------------------------------------------+------------+-------------------------------+
| query adapter (all trivial registrations) | 4.59 ms    | 3.65 ms: 1.26x faster (-20%)  |
+-------------------------------------------+------------+-------------------------------+
| contains (empty dict)                     | 136 ns     | 55.4 ns: 2.45x faster (-59%)  |
+-------------------------------------------+------------+-------------------------------+
| contains (populated dict)                 | 137 ns     | 55.0 ns: 2.49x faster (-60%)  |
+-------------------------------------------+------------+-------------------------------+
| contains (populated list)                 | 40.2 us    | 2.95 us: 13.62x faster (-93%) |
+-------------------------------------------+------------+-------------------------------+

Included benchmark numbers:

Current master, Python 3.8:

.....................
contains (empty dict): Mean +- std dev: 198 ns +- 5 ns
.....................
contains (populated dict): Mean +- std dev: 197 ns +- 6 ns
.....................
contains (populated list): Mean +- std dev: 53.1 us +- 1.2 us

This code:

.....................
contains (empty dict): Mean +- std dev: 77.9 ns +- 2.3 ns
.....................
contains (populated dict): Mean +- std dev: 78.4 ns +- 3.1 ns
.....................
contains (populated list): Mean +- std dev: 3.69 us +- 0.08 us

So anywhere from 2.5 to 15x faster. Not sure how that will translate to
larger benchmarks, but I'm hopeful.

It turns out that messing with ``__module__`` is nasty, tricky
business, especially when you do it from C. Everytime you define a new
subclass, the descriptors that you set get overridden by the type
machinery (PyType_Ready). I'm using a data descriptor and a meta class
right now to avoid that but I'm not super happy with that and would
like to find a better way. (At least, maybe the data part of the
descriptor isn't necessary?) It may be needed to move more code into
C, I don't want a slowdown accessing ``__module__`` either; copying
around the standard PyGetSet or PyMember descriptors isn't enough
because they don't work on the class object (so
``classImplements(InterfaceClass, IInterface)`` fails).

The DEFINE_STRING macro prevents the linter from seeing them as unused
so I temporarily redefined it to find all such variables.