Improve search integration with the documentation panel

[joguSD]

Previously, if you hit the input timeout while searching the documentation panel would close as input into the search buffer would be taken into consideration for the documentation to be shown. This PR does two things: 1) Ensures that the documentation panel only considers the prompt's input for show documentation 2) Will conditionally apply the search criteria against the prompt or documentation buffer based on which was previously focused before the search is activated

[joguSD]

Trying to write tests for this by leveraging the internal interfaces will be quite challenging. There's a vt100 terminal emulator called pyte that we could leverage to have the output of the shell go to an in memory string representation of the current terminal screen. Though, that only solves one problem and would likely also require a custom event loop that would allow us to provide input, inspect the current screen state, then provide more input, etc. I think it would be quite an effort to wire the prompt toolkit internals up in this way.

I also found hecate (by the same guy who wrote hypothesis) which is a testing framework designed for testing ncurses like terminal applications. It leverages tmux as a terminal emulator and interfaces with a tmux session to send input and determine the current screen state. It's relatively simple to write tests with it and I have one in this PR to test these changes. Unfortunately, actually running an instance of the shell to run tests against it currently is being blocked by having the documentation generate in the background. It would be better to have all the documentation already generated then run these kind of integration tests against them for consistency.

[joguSD]

I got a proof of concept done for the pyte version. It currently side steps the event loop issue by spawning it in another thread while retaining a reference to the input stream and pyte screen. Might need to do some more interface shuffling on the AWSShell class to ensure all environment sensitive things can be injected rather than assumed (such as history).

This approach has some pretty significant advantages over hecate: 1) No dependencies outside of python 2) A lot faster 3) Not as sensitive to timing issues 4) We control the AWSShell instance being run.