kv_list_keys will give you all the keys in the database.  Still TODO:
implement the "starting key" feature, so we can get a subset of keys.

Makefile.am fix for manual build linking issue. Bug fix in kvgroup-client for addr_str size argument to margo_addr_to_string. The size is in/out so need to be careful in the loop.

Split up "context" (the instance of the sds-keyval service) and
"database" (a specific collection of keys and values).

Furthermore, teach 'group' about this split.  Now a 'group' is a
collection of one context and one or more databases.

fixes #5 and #6
fully fixes #7

server-oriented code needs a few more libraries than client-side.
Should fix Matthieu's build problem where boost was not propigated up to
mobject.

one purpose of the alternate backends is to get us building when we
don't have gcc5 or newer.  If we always build bwtree, then we're not
helping things.  Now BwTree is only built if it's the selected backend

C compilers on bebop are still c99 by default.

bebop machine (centos 7) has automake 1.13.4 by default.  Should be good
enough for us.

Fix BerkeleyDB interface implementation to be multithread tolerant. Added simple ABT_mutex in kv-server to protect access to the 2 datastore related global variables in open_handler. Successfully tested with ParSplice.

split up header into public and private/internal versions.

Fixes: #4

populaing a "kv_config.h" makes it easier to inspect what configure
thinks it found vs. looking through compiler command lines.

since BwTree is an optional backend, the BwTree flags should be selected
at configure time

these small wrappers around margo init/finalize aren't helpful when
compositing with multiple services

also lets us use strange/specific CXXFLAGS only when needed

only the spack package provides a pkgconfig file.  for other leveldb
installations we'll have to check for headers ourselves.

populaing a "kv_config.h" makes it easier to inspect what configure
thinks it found vs. looking through compiler command lines.