Upgrade HtmlUnit to v2.18 (from forked version)

* master:
  commons-compress -> 1.10
  Noting #1788
  Remove hardcode in tar test
  [FIXED JENKINS-10629] Unbroke stream with new tar implementation.
  Revert "Revert "FIXED JENKINS-10629] - Enable BigNumber mode to support archiving of files with size >8Gb""
  Revert "Revert "[JENKINS-10629] - Migrate the Tar archives handling code to commons-compress""
  Fix at-since from PR #1788
  Merge PR #1788: Make plugin manager pluggable

Causes bytes lost and truncated tar archive.
TarBuffer not used in TarArchiveOutputStream.

Revert "Revert "FIXED JENKINS-10629] - Enable BigNumber mode to support archiving of files with size >8Gb""

This reverts commit ee573009.

This reverts commit 8f1280a8.

Squashed commit of the following:

commit f2c6ad272fe9bd3b93e21001be63694f2b8841ec
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Thu Aug 20 18:28:09 2015 -0400

    After much discussion, just remove the helper method to let a plugin manager override easily get instance of the PluginManager

commit a96f933d8dcd1bf2f02b216ccf0a6517bb649121
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Thu Aug 20 16:59:07 2015 -0400

    Add javadocs and restrict access on the PluginManagerStaplerOverride getManager method

commit 7e1facf887ab1b3219b116cac42a785c99b2a635
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Thu Aug 20 12:50:47 2015 -0400

    Simplify plugin manager test using @TestExtension

commit b9f2841f6d3453a2b9a38ec3d90247ce8020add3
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Thu Aug 20 10:32:49 2015 -0400

    Modify the Since version for pluginmanager overrides

commit 3af45ce579310f157f4cc3ee128a14a33bd4e480
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Thu Aug 20 10:28:14 2015 -0400

    Add test for PluginManagerStaplerOverride

commit 786409456590026cb2c4f56f687bae71d05990d8
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Mon Aug 17 20:59:33 2015 -0400

    Changes from review, mostly comments and rename of class

commit 1c23f086eced9fb46a7c643174a50f71bc64ec37
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Mon Aug 17 12:58:07 2015 -0400

    Cleanup for PluginManager extensibility/UI overrides

commit e105a13bcf9324c87d896ddc04ae03f905dd0f0a
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Fri Aug 14 15:23:07 2015 -0400

    Different approach to plugin manager UX pluggability, using Extension and StaplerOverridable

commit 81f21fba08a170a0f11bb88b769d6b8c69e4d1c0
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Thu Aug 13 21:32:45 2015 -0400

    Initial round of changes to allow  overriding PluginManager

    This includes renaming the setTarget method to setUIProxy and defining an interface to retain APIs
    Fields have not been mapped through yet as that is still in progress
    Also, an AbstractStaplerUIProxy has been defined to delegate Stapler bound calls to the pluginManager

commit eeef666de718ddfca04895108f5015eb13729e5e
Author: Sam Van Oort <samvanoort@gmail.com>
Date:   Wed Aug 5 17:40:50 2015 -0400

    Make plugin manager pluggable

Lots of tests di "stuff" that relies on JS execution and then check for a result in the DOM by querying it. We should check make sure all JS has finished executing before we perform the DOM query.

runScriptOnOfflineComputer has always been dodgy from what I can see. It was relying on the slave not being connected at the time the doScript call happened. I tested this on a vanilla master and it's easy to make this test fail  -just put a loop in after the slave is created, waiting on the slave channel to be connected before making the doScript call.

This change makes sure the slave has connected and disconnected before continuing.