Fix partial updates in Python using dicts

[sc1f]

Fixes #1268 by refactoring the data loading path for dict-shaped datasets.

There are some inconsistencies between how None values are handled in JS and in Python. In JS, null and undefined in a dataset mean different things in a partial update:

table = perspective.table({a: [1, 2, 3, 4], b: ["a", "b", "c", "d"], c: [10, 20, 30, 40]}, {index: "a"})
table.update({a: [3], c: [100]}) // "b" is `undefined` here, so values in `b` are not modified.

undefined resolves to a no-op, so values that are undefined will not be modified or overwritten. However, if the value is replaced with null:

table.update({a: [3], b: [null], c: [100]}) // "b" is explicitly "null", which will unset the value at "b" for pkey `3` and set it to null.

To remedy this, the Python library uses a method called _has_column, which ascertains whether a given row contains a given column name. If the column name does not exist, it is a no-op similar to undefined, but if the column name exists then it will be overwritten with the new value.

Thus, a row-oriented update of [{a: undefined}] that would work as a no-op update in Javascript does not work in Python. In the column-oriented case, #1268 illustrates the issue where a missing column in a columnar dataset would be treated as an overwrite, and not a no-op. This PR fixes the behavior by treating missing columns in columnar datasets as no-ops, and has been tested. This behavior is now equivalent between JS and Python.

There will always remain Python-specific idiosyncrasies around partial updates. For example, this update works in JS:

table.update({
  a: [2, 3, 4],
  b: [null, undefined, null] // delete, no-op, delete
})

but it would not work in Python, because one cannot satisfy the "all columns must have the same # of rows" requirement AND mark a column[row] as a no-op, and there is no way to reconcile this behavior:

table.update({
  "a": [2, 3, 4],
  "b": [None, None, None] # delete, delete, delete
})

[texodus]

Looks good! Thanks for the PR!