Following up on a discussion from last month [1], I am reviving my
proposal from a couple years ago [2] to improve ob-python results
handling. Since it's a relatively large change, I am sending it to the
list for review before applying the patch.

The patch changes how ob-python handles the following types of
results:

- Dictionaries
- Numpy arrays
- Pandas dataframes and series
- Matplotlib figures

Starting with dicts: these are no longer mangled. The current behavior
(before patch) is like so:

#+begin_src python
  return {"a": 1, "b": 2}
#+end_src

#+RESULTS:
| a | : | 1 | b | : | 2 |

But after the patch they appear like so:

#+begin_src python
  return {"a": 1, "b": 2}
#+end_src

#+RESULTS:
: {'a': 1, 'b': 2}

Next, for numpy arrays and pandas dataframes/series: these are
converted to tables, for example:

#+begin_src python
  import pandas as pd
  import numpy as np

  return pd.DataFrame(np.array([[1,2,3],[4,5,6]]),
                      columns=['a','b','c'])
#+end_src

#+RESULTS:
|   | a | b | c |
|---+---+---+---|
| 0 | 1 | 2 | 3 |
| 1 | 4 | 5 | 6 |

To avoid conversion, you can specify "raw", "verbatim", "scalar", or
"output" in the ":results" header argument.

Finally, for plots: ob-python now supports ":results graphics" header
arg. The behavior depends on whether using output or value
results. For output results, the current figure (pyplot.gcf) is
cleared before evaluating, then the result saved. For value results,
the block is expected to return a matplotlib Figure, which is
saved. To set the figure size, do it from within Python.

Here is an example of how to plot:

#+begin_src python :results output graphics file :file boxplot.svg
  import matplotlib.pyplot as plt
  import seaborn as sns
  plt.figure(figsize=(5, 5))
  tips = sns.load_dataset("tips")
  sns.boxplot(x="day", y="tip", data=tips)
#+end_src

Compared to the original version of this patch [2], I tried to
simplify and streamline things as much as possible, since this is a
relatively large and complex change. For example, the handling for
dict objects is much more simplistic now. And there are other
miscellaneous changes to the code structure which I hope improve the
clarity a bit.

[1] https://list.orgmode.org/CAOQTW-N9rE7fDRM1APMO8X5LRZmJfn_ZjhT3rvaF4X+s5M_jZw@mail.gmail.com/
[2] https://list.orgmode.org/87eenpfe77.fsf@gmail.com/