Multiple, conflicting, index memory release attempts may lead to segfault

Created by: avullo

Some glitches in the way index memory is released.

From @jmarshall:

"Running the t/03cram.t tests under valgrind shows a double-free bug (the cram index is freed via both hts_close and hts_idx_destroy), present also on the master branch and with any version of HTSlib, that occasionally leads to a segfault (only once in my attempts to reproduce this). I suspect it is this that caused the failed travis run, which ran the t/03cram.t tests twice: once successfully (via ./Build test) and once with a segfault (via travisci/harness.sh)."