This reverts commit e2168722.

- Use the proper block cipher mode.
  Or else the information about the plain text still ends up revealing as a pattern without the attacker knowing the key.
- No need to hide SLAVE_SECRET from the encrypted payload.
  jnlpMac is needed to decrypt this payload to begin with, so there's no point in hiding it. This simplifies the code a little bit.
- Using a newer slave installer that uses the -secret option

Jesse's original patch

This came from pull request #705

Fixed the HTTP request thread saturation problem with Winstone.

eval("") is no-op in IE but window.execScript("") isn't.
So bring the uniformity outside

Failure to load JNA should be handled graciously

[FIXED JENKINS-16776] surefire-reports not detected for android-maven-plugin

(Or should we just check the status code?)

translation cleanup

purged duplicated translations for GlobalCrumbIssuerConfiguration
("Prevent Cross Site Request Forgery exploits").
Changed property name "Crumb" to "Crumbs" because this is the value that the existing translations actually use.

Improve UpdateSiteTest by adding valid Update sites

Removing old static field, which was not friendly to tests, especially multithreaded runners.
Cleaned up existing tests to use updateDirectly rather than POSTing raw JSON.

[JENKINS-15156] AbstractProject.builds accessed by Maven & matrix module builds before onLoad or onCreatedFromScratch called.
Seem to need to keep the deprecated no-arg RunMap initializer though it is unclear to me how it could work.

so iterate on itemGroup.getItems and filter by job names