diagnostics should show less aggressively when typing code in an editor

At least in Python, we currently have a very happy trigger finger with showing diagnostics (both in the console and in the source editor, though we're dealing with the console in a separate issue #1114). For example, if you simply enter this:

>>> x = 

You will get a red underline indicating an "error" in your code:

Especially for less experienced developers I think this will be at best annoying and at worst provoke users to actually figure out "what's wrong" with the code they just entered.

I think we need a fancier set of heuristics for when to show diagnostics around active edits. One thing I've seen work well is to never show diagnostics on the line where the cursor currently is. We could also never show diagnostics if the current line is empty. Another thing we could do is have diagnostics run debounced over a longer period (e.g. 3 seconds). The goal is to never mis-detect and incomplete edit as an error or warning.

We'll need to play around with different strategies in order to get this just right.

Reproduce the Python x = example in the source editor and inspect the diagnostic behavior during active edits; the issue names no implementation file or test. Done means agreeing on and validating heuristics that avoid treating incomplete edits as errors, including the cursor line or empty lines, with any debounce behavior tested against the editor experience.