Pemi

------

* Adds PemiTabular methods that will create an HTML file when tests fail. This can
help with debugging failed tests.

------

* **BREAKING** Allows one to make more natural connections to nested pipes. The sources
of a pipe can now be connected to the sources of nested pipes and the targets of
nested pipes can now be connected to the targets of the parent pipe. This will break
code that has workarounds for this issue.
* Prevents Pandas from raising a warning when concatenating CSV files

--------
* pemi.transforms.isblank now treats pandas.NaT as blank

------
* pemi.transforms.isblank no longer raises an error when given an array or dict with
more than 1 element.

------
* Allow CSV and SqlAlchemy pipes to work without defining a Pemi schema

------
* Removed ``fake_with`` option when constructing test data tables. Custom data fakers should
now be specified in schema field metadata field ``faker``, which expects a function that
when called will return faked data.
* Replaced internal pd_mapper with pandas-mapper package. Users will have to migrate
to pandas-mapper if they were using PdMapper/PdMap functionality in Pemi.
* Removed support for ``allow_null`` metadata. Never used in production.
* Upgraded many package dependencies.
* Removed various unused features.