And for check_LTLIBRARIES to noinst_LTLIBRARIES too.

This is so a regular "make" will build the test programs and
libraries too.  So the "make check" will just run the tests (and
not build them).  This is to help with test-driven development.

This reverses some from #1091, plus does the rest too.

Note that doing a "make check" before a "make" does not currently
work due to dependencies depending on each other (e.g. python/isc/log
tests needs datasrc but python/isc/datasrc tests need log_config_update).

I mentioned this on jabber last week:
Why would it want to do the local directory tests before the testdata
is generated?

This issue is never noticed yet because the testdata was built
at build time and not check time.

use RTLD_GLOBAL when opening modules, otherwise the exceptions some methods raise may not be recognized in catches

Conflicts:
	src/lib/datasrc/sqlite3_accessor.cc
	src/lib/python/isc/datasrc/client_python.cc

this would have revealed the DataSourceClient_getUpdater bug.

DataSourceClient.get_updater must return None when it cannot find the zone.
Otherwise the returned updater would be incomplete and would subseuqently
cause a disruption (such as program crash).

its counter intuitive behavior and surprising results rarely justify
the introduction, and, in fact, it caused a regression:
http://git.bind10.isc.org/~tester/builder/BIND10/20111007175000-Solaris10-sparc-Sunstudio/logs/unittests.out
Apparently the reason why this was introduced is to assist the comparison
tests used in a dhcp6 test.  To work this around I replaced the test code
using toText().  The right solution is to introduce operator<< for IOAddress,
which will satisfy gtest (we already have operator== so EXPECT_EQ should
work otherwise).