Remove a race hazard from the execrable Support_TestWebServer

Use of this class was resulting in threads being left in a
spinning state due to the accept socket being closed but
"running" still being true: the exception thrown from
Socket.accept() was being swallowed.

This is likely because there is a race between the thread
actually starting (i.e. run() actually executing)
and, in short-lived tests, the server being shutdown:
if AcceptThread.close() was called before AcceptThread.run()
then when run() actually executed it would loop forever.

Some dead code has been removed.

Test: Ran the CTS tests
Bug: 29820565
Bug: 29365528
(cherry picked from commit 5558171b)

Change-Id: I895ecbd171c9495ddda85de740519ba1b2c51d62