* (bug 29174) Fix regression in upload-by-URL: files larger than PHP memory limit work again
r65152 switched upload-by-URL ($wgAllowCopyUploads) to use Http / MwHttpRequest class instead of CURL directly.
While this is mostly nice, it switched from saving large files directly to a temp file to buffering them in memory, causing large files to fail when they hit the PHP memory limit.
Fix uses MwHttpRequest's callback capability to override the read handler; now appending it to the temporary file as we go, and can confirm that largish files work again; was able to upload a 64mb .ogv that previously didn't work for me: http://prototype.wikimedia.org/tmh/images/b/b2/File-Arborophila_brunneopectus_pair_feeding_-_Kaeng_Krachan.ogv
Also expanded the documentation on MwHttpRequest::setCallback to clarify the function parameters and return value for the callback (which currently matches the low-level CURL handler's callback directly).
Note that the non-CURL implementation doesn't abort the read if the callback doesn't return the expected number of bytes, but this is an immediate fatal end of request on the Curl backend. May want further cleanup.