Remove containers whose contents are never read (java/unused-container CodeQL alerts)

[vharseko]

Clears seven java/unused-container alerts (#688, #689, #690, #691, #693, #694, #695). No behaviour changes.

Removed

Alert	Container	Why it is safe
#695	ErrorLogAccountStatusNotificationHandler.NOTIFICATION_TYPE_NAMES	Filled by a static initialiser, never read anywhere.
#694	NodeSearcherQueue.cancelList	Had put/remove calls but no reads. The cancellation itself is done by task.cancel(); this map was bookkeeping nobody consulted.
#693	GenerateMessageFileMojo.CATEGORY_DESCRIPTIONS	24 put calls, zero reads. The LocalizableMessage import became unused and was dropped with it.
#691	OnDiskMergeImporter.importedBaseDNs	Populated inside the phase-two loop, never read.
#690	NewIndexPanel.configurationAttrNames	See below.
#689	PropertyValueEditor.addedValues	Computed and discarded, together with the previousValues set that only existed to feed it.
#688	DSConfig.getPropHandlers	The only one of the five handler maps that is not passed to SubMenuCallback. The surrounding loop is kept, because it also populates the relations set.

A note on NewIndexPanel

The panel classified attributes three ways — standard, configuration, custom — but only rendered the custom and standard categories, so configuration attributes were silently dropped into a set nobody read.

That looks like an oversight at first glance, but AbstractVLVIndexPanel does exactly the same thing: same three-way split, same two categories in the combo box. Two independent panels behaving identically reads as intent — configuration attributes are not offered for indexing — so this change does not add a third category. It only expresses the existing exclusion directly:

else if (!Utilities.isConfiguration(attr))
{
    // Configuration attributes are not offered for indexing.
    customAttrNames.add(name);
}

If the intent was in fact to offer them, that is a separate, user-visible change and should be its own PR.

Two alerts deliberately left open

#686 — SubCommandArgumentParser.globalArgumentMap (java/empty-container) is not dead code, it is unfinished code: the map is read in two places but nothing is ever put into it.

• hasGlobalArgument(name) always returns false, so the conflict check in SubCommand.addArgument() never fires and a subcommand argument can silently shadow a global one.
• The duplicate check in addGlobalArgument() never fires either. globalLongIDMap covers part of it, but only on the !longArgumentsCaseSensitive() path.

Fixing it — populating the map, or pointing both reads at globalLongIDMap — switches on two validations that are currently dead. If any tool (dsconfig, base64, ...) has such a name clash today, it would start failing at startup with an ArgumentException. That cannot be settled by reading the code; it needs the CLI tools to be exercised. It deserves its own change rather than riding along with a cleanup.

#692 — InternalConnection.listeners: addConnectionEventListener()/removeConnectionEventListener() work, but nothing ever iterates the list, so internal connections never notify their listeners of a close, an error or an unsolicited notification. The field cannot simply be deleted without breaking the public Connection contract, and wiring up the notifications is a feature, not a cleanup.

Testing

• mvn -pl opendj-config,opendj-doc-maven-plugin,opendj-server-legacy compile — BUILD SUCCESS
• mvn -pl opendj-config test — 544 tests run, 0 failures, 0 errors