Refactor Quick Add to require fewer clicks and reduce boilerplate

This is another somewhat monolithic change to the frontend, to rewrite the Quick Add component that appears at the bottom of the table of contents.

Includes the following issues:

1968
1975
1976
1977

Before

Status quo is the user enters a case name in the Quick Add, then a temporary item is added, then they click "Specify Document", then they click "Search", then they pick the results (they have to click directly on the title, not the row), then the item is changed in place.

https://user-images.githubusercontent.com/19571/230987676-05cdcfad-0c69-40e1-b71c-7ac311c07ee4.mov

After

Now the search form is embedded in the Quick Add, so when either automatically (by a matching regular expression) or manually (by pulling the dropdown to Legal Documents), the search happens inline, and the matching item is inserted in place. This reduces the number of clicks from at least 4 to 2.

https://user-images.githubusercontent.com/19571/230987720-74c6b7d3-8395-4bb1-aada-612c0fb38a70.mov

The other functionality is unchanged: users can still paste links from within H2O to auto-clone, paste whole outlines, or add sections or custom content as before.

Under the hood changes

Advanced search from Add Content has been refactored into a discrete component and is used in both places. Similarly the actual search API call is now its own function.
Dropped unnecessary external libraries like lodash and Axios in favor of native browser calls.
QuickAdd now has some unit tests (it had none before)
Before, the Clone and Outline options were implicit and then mutated the resource type model in confusing ways. They are now first-class citizens which makes user-facing documentation more straightforward.
Dropped some dead code (unWait()?)

@cath9 and I will probably iterate on this on staging before it gets to prod.

harvard-lil / h2o