[frmoretto]

Fixes #768.

parse() called propertyStream() on array nodes, which Jackson returns empty
for anything that is not an object, so URIs inside an array held as an object
property value were never extracted. Those array elements now go through the same
value handling as object values, which is also what lets a bare URI string inside
an array be extracted and not only an object holding one.

A known limit this does not change: parse() still starts from
rootNode.properties(), so a document whose root is itself an array yields no
links.

valueStream() rather than elements(): both iterate the elements of an array,
and neither is deprecated in the 2.22.2 Jackson this builds against, so the
choice is one of style — valueStream() matches the 2.19-style node API this
file already uses, properties() and the propertyStream() being replaced here.

The extractor already recurses through nested objects without a depth guard.
This patch applies the same recursive approach to arrays; adding a depth limit
is outside this defect fix.

ExtractorJsonTest covers it: the object-property case passes before and after
as a control, and the two array cases go from failing to passing. Full modules
suite, JDK 17 on Ubuntu 22.04 with LANG=C.UTF-8: 254 run, 2 failures, 1 skipped
before the change; 254 run, 0 failures, 1 skipped after. The single skip
(FetchHTTPTest) is the same in both runs.

Built and measured against abd17fa.

[ato]

Thanks!