Proposal to add hts_log callback function and hts_log_set_logger()

[jmarshall]

A somewhat tidied-up version of @lczech's concrete proposal in #1397 for overriding the default printing of hts_log messages to stderr. Draft, as the new function (and associated typedef) would require documenting in hts_log.h.

I don't know whether the hts_log_* functions are called from threads that HTSlib has started. If so, this callback function would be called from those threads — which might be “interesting”. This PR doesn't consider what to do in this case.

[daviesrob]

I think this looks OK in principle, although we should do something about the (not unlikely) possibility of the hts_logger() callback being called from a thread. I think at the least, this should be mentioned in the callback documentation. It would also be a good idea to add a mutex, to be held both when setting the callback function and when calling it. That would both eliminate the small risk of data races, and allow us to guarantee that the callback will not be called simultaneously from different threads.

[jmarshall]

@lczech or the maintainers should feel free to do so, and may find this PR (or just the commit) a useful starting point.