Fix a segfault on global pool cleanup after merge which has been

Fix a segfault on global pool cleanup after merge which has been
occurring on some platforms since r28442.  It appears that, due to the
ne_decompress destruction in the request cleanup handler,
svn_ra_neon__parsed_request must always be called with a pool whose
lifetime is no longer than the RA session pool's.  (This will be
documented in a future revision.  (Update: actually, r29231 fixes
this in a different way.))

* subversion/libsvn_ra_neon/get_location_segments.c
  (svn_ra_neon__get_location_segments): Run everything (including
   svn_ra_neon__parsed_request) in a subpool.

git-svn-id: http://svn.collab.net/repos/svn/trunk@29230 612f8ebc-c883-4be0-9ee0-a4e9ef946e3a