Clarify /etc/foo.d pattern

In the past years more and more software has been changed to use /etc/foo.d
patterns.

However the correct way to create an /etc/foo.d root have never been documented
anywhere, different applications use slightly different conventions, they all
rediscover the same problems and it's a huge mess for sysadmins.

The FHS should clearly state how everyone is supposed to structure a clean
/etc/foo.d. I propose that at the minimum the following conventions be stated:

1. extension: an application should only parse files with a pre-defined
extention in its /etc/foo.d

 Rationale: many utilities will create temporary (or not) copies of those files
when they are edited (.swp, .bak, .rpmsave, .rpmnew…) breaking software that
parses all the files in foo.d

2. ordering: if the files in foo.d need to be parsed in a specific order, they
should be prefixed with number- and parsed in ascending number prefix order

3. snippets: if a project wishes to make available optionnal pre-composed
configuration snippets, they should be installed in
/usr/share/something/foo.avail, and activated by symlinking from this directory
to foo.d. If the sysadmin wishes to derive one of those files he should copy it
to /etc/foo.d and edit the copy only