Refactor Jupyter Notebook support

[fdxmw]

PyRTL's Jupyter Notebook support is quite messy, see _currently_in_jupyter_notebook() and its callsites. There are several problems:

1. _currently_in_jupyter_notebook() does not work reliably across different notebook implementations.
2. It is confusing that PyRTL functions like render_trace() behave differently depending on whether they're running in a notebook environment or not, and even more confusing when we can't reliably detect notebook environments.
3. PyRTL modules often mix data and how that data is presented. See Block.__str__() for example. This approach tends to spray UI code across core PyRTL modules, for example every __str__() method would end up with tons of if statements to support all presentation styles.

To clean this up:

1. Refactor PyRTL's core modules so they only return data, so the user can do whatever they want with that data (pass it to another library, throw it in a Protocol Buffer, etc).
2. All presentation and formatting code should move into separate modules built on top of these core-data modules.
3. Most of these presentation-and-formatting modules should not be top-level components in the pyrtl module.
4. Users should invoke these formatting modules explicitly, for predictable behavior.
5. Avoid magic code that guesses what the user wants to see. Magic code inevitably does the wrong thing for someone at at some point. Magic code is a maintenance burden because it usually can't be fixed, only slightly improved by piling on more heuristics.

Edit: Tim and I discussed this a bit more offline. Two more notes:

6. Users shouldn't have to repeat themselves, so they shouldn't need keep reminding us about how they want their output styled, for example. NumPy's np.setprintoptions seems like a good example of designing with this principle in mind.
7. Some PyRTL functions like render_trace can produce a lot of output. These functions must support writing the output to a file, without buffering the entire output in memory.