Allow more functionality through system METIS

[acxz]

This is a follow up PR to #570 as when it was merged I was not available to provide my review.

There is no reason to specify the full local metis.h include path, with how metis is treated in cmake, metis.h works for both local and system metis.

See: https://github.com/borglab/gtsam/pull/570/files#diff-0106693c254359c8aceceb57d5b6c0643760cc8cce6760ccfd2d31bc6d0be2dbR26
where only metis.h is used without the ifdef.

More importantly this PR enables metis's internal api ("metislib.h") via extern C for SYSTEM_METIS. We already use the internal API for our local metis install and since we carry the header files of metis with us I do not see a problem in enabling extern "C" for system metis as well. The only problem that can occur with this is a desync between the local headers we carry for metis and what metis is compiled under at the system level. Considering metis is a very old and stable package that hasn't seen updates for a long time, I do not see this issue occurring and if it does should be easily fixable, by just updating our local metis copy.

Tagging @jlblancoc and @varunagrawal for reviews. Also @berndpfrommer ig

[dellaert]

What specific issue are we fixing? Was there a problem that prompted this PR? Is there an issue associated with the problem?

[acxz]

What specific issue are we fixing?

Unnecessary ifdef when including <metis.h>
FindSeparator tests not run under system metis

Was there a problem that prompted this PR?

From my review comment in the last METIS PR:

Even if that is changed, system metis does not provide metislib.h which means we won't be able to use the functionality of it. Thus rendering the SYSTEM_METIS flag broken. If its going to be broken when merged in, we should just not merge it in.

Currently the ifdef guard for SYSTEM_METIS just disables the associated test. We should fix that by allowing tests to run under system metis.

Is there an issue associated with the problem?

Here is the issue: #292 (comment)
This PR implements @jlblancoc 's approach in #292 (comment)

[dellaert]

Awesome. Requesting José as a reviewer, who can merge after approving.

[varunagrawal]

Minor nits but LGTM