PDF retrieval doesn't work if DOI/URL page is generated via JavaScript

[dstillman]

E.g., https://ieeexplore.ieee.org/document/6896636 (reported here)

IEEE pages are generated client-side, but we're using HTTP.request() (i.e., XHR) to load the page in Zotero.Attachments.downloadFirstAvailableFile().

The general solution here would be to use a hidden browser from loadDocuments() instead, but the DOI/URL page could be a PDF document itself, and loading a PDF document in a hidden browser can cause it to either show the open/save dialog or automatically open in an external PDF reader (depending on whether the user has gotten the open/save dialog before). We'd have to figure out how to gracefully handle non-native responses — perhaps by automatically saving all non-native types to a temp file and running a callback, since we never actually want to show an open/save dialog.