Pjpersist

------------------

- Fixed an edge case when the serializer gets a mapping with a key `dict_data`.
Reading such object failed.

- Fixed an edge case with the serializer, when an object's state living
in a persistent object became 'empty'. Basically the state was just
`{'_py_persistent_type': 'SomeClass'}`
`SomeClass.__setstate__` was not called, thus the object could miss
attributes. Like a subclass of `UserDict` would miss the `data` attribute.

- Removed checking for 0x00 chars in dict keys. Turns out PostGreSQL just
can not store 0x00.

------------------

- Support for sub-second datetime and time resolution during serialization.

- Add `use_cache` argument to `PJContainer._load_one()` to support ignoring
the cache. (This became handy if a container keeps track of multiple
versions of an item and you try to load all old revisions.)

------------------

- Make `id` and `data` column name configurable via `_pj_id_column` and
`_pj_data_column` attributes in `PJContainer`, respectively.

- Auto-assign a name to objects when using `PJContainer`, not just
`IdNamesPJContainer`.

------------------

- Support for Python 3.7. Removed Python 3.5 testing from tox.

------------------

- No need to log in tpc_finish.

------------------

- Implemented skipping tpc_prepare when DM has no writes.
We found out that AWS Aurora is dog slow at the moment on tpc_prepare.
When the DataManager has no writes, there's no need to call tpc_prepare.
See `CALL_TPC_PREPARE_ON_NO_WRITE_TRANSACTION`, by default True for backwards
compatibility.

- Added ability to log whether the transaction had writes.
See `LOG_READ_WRITE_TRANSACTION`, by default False