RFR: JDK-8222791 Order of evaluation of -link params in Javadoc tool reversed -> regression with split packages

[Jonathan Gibbons]

Please review a simple fix and a corresponding test for a regression in the
way that a series of -link or -linkoffline  options are processed. When
a package appears in multiple APIs, either as different versions, or for 
a split
package, the behavior changed from "first one wins" to "last one wins".

The problem and basic solution are given by Uwe in the JBS issue. 
Thanks, Uwe.

The fix here is as outlined in the JBS issue (replace "put" by 
"putIfAbsent"),
together with some localized cosmetic changes.

The test creates a couple of libraries and places them in different order
on the javadoc command line and verifies that "first one wins". The 
libraries
are created in two variants, using "package-list" and "element-list".

As with any good regression test, the test fails on an unmodified JDK.

-- Jon

JBS: https://bugs.openjdk.java.net/browse/JDK-8222791
Webrev: http://cr.openjdk.java.net/~jjg/8222791/webrev.00/