[pyobj-] improve display of sets and simple objects

[midichef]

This set of commits is aimed at making pyobj-expr more user-friendly. As it stands now, the results from looking at certain values are quite strange.
None: gives a sheet with no rows
set([1,2,3]): a sheet with no rows (because what is shown are its attributes, not its elements)
str(): a sheet with no rows, with the header text, which is different from the previous two cases
1 or True: a sheet of attributes, hard to make sense of:

attribute   value docstring
denominator 1     <type int>
imag        0     <type int>
numerator   1     <type int>
real        1     <type int>

I've added a new sheet type, PythonAtomSheet to hold None or a bool, a number, or an empty string. The goal is to improve the display. The change makes it easy to distinguish an expression value of None from "", due to the ⌀ shown. And it shows bools/numbers in a simple 1x1 cell, instead of a 4x3 sheet of display attributes.

The new sheet also doesn't allow diving deeper into such simple objects. By contrast, the way visidata already works is, after pyobj-expr for a number 1, you can keep hitting Enter to dive in on the denominator field infinitely. It's quite confusing for a new user.

And I fixed a bug where if a sheet had no rows, open-row takes the user to a new PyobjExpr sheet.
(to reproduce, vd nonexistent_file.tsv then ENTER)

Some quirks of this PR:

1. None is shown as it is in a standard sheet, an empty cell with ⌀ next to it. Maybe it should show the text "None".
2. If you view bools, the call to deduceType will put ? next to the column header.
3. The empty string goes into PythonAtomSheet unlike regular strings that go into a TextSheet. I did that to prevent further dives into the empty string. But that creates an inconsistency, the header for empty strings is "value", but for all other strings it's "text". To get rid of the special handling for empty strings, just change the code to:

        if pyobj is None or isinstance(pyobj, numbers.Number):
...
        elif isinstance(pyobj, str):
            return TextSheet(*names, source=pyobj.splitlines() if pyobj else [''])
        elif isinstance(pyobj, bytes):
            s = pyobj.decode(cls.options.encoding).splitlines()
            return TextSheet(*names, source=s if s else [''])

[saulpw]

We're heading more towards using prettykeys like Enter for keybindings, and deprecating the ENTER variable (though we haven't officially done so yet). So let's not change these instances.

[saulpw]

Let's revert the changes to use ENTER for the keybinding, and then this is good to merge. Thanks @midichef!

[midichef]

Okay, I reverted the ENTER change, then consolidated the others. This PR is done.