Convert libnbcompat to regular buildlink3

[Jonathan Perkin <jperkin%joyent.com@localhost>]

Currently pkgtools/libnbcompat is implemented in such a way that every
package that requires it will build its own private copy of the static
library and pull it in.

This is a waste of time and resources, and during a --full bootstrap
means we build 5 identical copies of libnbcompat.

I have a diff which converts it to a regular buildlink3 dependency,
and patches packages to pull it in as a BUILD_DEPENDS and find the
installed headers instead of the inplace ones:

  https://github.com/joyent/pkgsrc/commit/61ab01a01e32a27ceeba3e07dee4b66839d9535d

On SmartOS with "./bootstrap --full" this takes us down from 5 builds
to 2, and the time to bootstrap from:

  real     8:38.8
  user     4:16.3
  sys      2:48.0

to this:

  real     6:49.6
  user     3:27.5
  sys      2:09.4

This also helps NetBSD bootstrap.  A default "./bootstrap" on NetBSD
6.1.5 will build 3 copies:

  /tmp/pkgsrc/libnbcompat
  /tmp/pkgsrc/wrk/pkgtools/pkg_install/work/libnbcompat
  /tmp/pkgsrc/wrk/devel/bmake/work/libnbcompat

taking:

  48.54 real        27.00 user        14.43 sys

whereas with the patch applied it builds only two:

  /tmp/pkgsrc/libnbcompat
  /tmp/pkgsrc/wrk/pkgtools/libnbcompat
  
and takes:

  42.95 real        23.36 user        12.74 sys

I have tested this in a full bulkbuild here:

  https://us-east.manta.joyent.com/pkgsrc/public/reports/upstream-bulktest/20150708.1604/meta/report.html

and cannot find any fallout, other than existing nbcompat-related
failures which I will be fixing shortly.

Two possible issues with the patch that I can think of are:

 - We have to add pkgtools/libnbcompat as an exclusion for
   USE_CWRAPPERS=yes to avoid circular dependencies.  Maybe we want to
   add a package-settable variable to disable cwrappers rather than
   hardcoding PKGPATHs in bsd.prefs.mk.

 - It removes the nascent support for LIBNBCOMPAT_USE_PIC which tnn@
   introduced 7 years ago.  That work, as far as I can see, was never
   finished anyway, and there's no particular reason why my work would
   prevent a similar feature to be introduced in the future.

Please review.

Thanks,

-- 
Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com