IndexFetcher: skip opening a searcher when the CoreContainer is shutting down

A polling fetch that finishes just as the node shuts down calls CoreContainer.getCore(), taking a fresh reference to a core the container is already tearing down. The container's own close then can't complete, and the last decref lands on the indexFetcher thread — so that thread ends up running the whole SolrCore.doClose(). If IndexWriter.close() fails there (the fetch has just swapped index directories out from under it), directory refcounts are left dangling and CachingDirectoryFactory.close() blocks for its full 12s wait before giving up on them.

Checking CoreContainer.isShutDown() first avoids taking the reference at all; opening a searcher at that point would fail anyway. This narrows the window rather than closing it — shutdown can still begin right after the check — but it covers the common case, since CoreContainer.shutdown() sets the flag before it starts closing cores.

Found while investigating an intermittent TestUserManagedReplicationWithAuth failure, where the aborted doClose() never reached its ObjectReleaseTracker release and 49 objects were reported leaked.