Adopt upstream XML trailing-comment formatting

[timtebeek]

Fixes the main build failure in https://github.com/openrewrite/rewrite-migrate-java/actions/runs/30564092748 (22 tests across two classes).

Depends on openrewrite/rewrite#8355, now merged and published to 8.89.0-SNAPSHOT.

Cause

openrewrite/rewrite#8353 made XML auto-format keep a comment on the same line as the element it trails, instead of reflowing it onto its own line. Both failing test classes relied on that reflow.

AddMockitoJavaAgentToMavenSurefirePlugin

The recipe builds <configuration><!--suppress MavenModelInspection --><argLine>…</argLine></configuration>. Where there is no existing <configuration> the comment is the first content of its tag and still reflows, so those cases never failed. Where the <argLine> is appended into an existing <configuration>, the comment is no longer first content and stayed glued to the preceding sibling:

-  </systemPropertyVariables>
-  <!--suppress MavenModelInspection -->
+  </systemPropertyVariables><!--suppress MavenModelInspection -->
   <argLine>…</argLine>

which reads as suppressing an inspection on <systemPropertyVariables> rather than on the <argLine> it is meant for. Seeding the comment with a newline prefix keeps it on its own line in both paths.

JpaCacheProperties

The 20 failures here split in two, and only the first half is a formatting bug.

13 cases — comment orphaned by removal. The recipe deletes a <property/> that had a trailing comment; the deleted element's prefix carried the line break, so the comment had none left and collapsed onto the previous line:

-  <!-- Connection properties -->
-  <!-- remove -->
+  <!-- Connection properties --><!-- remove -->

openrewrite/rewrite#8355 fixed exactly this — but in RemoveContentVisitor, whereas this recipe removed via filterTagChildren, which drops content outright and never saw the fix. Both call sites remove a single known element, so RemoveContentVisitor is the natural fit; switching them over restores the original expectations for these 13 with no test changes.

7 cases — nothing removed. Here the element the comment trails is retained (<!-- leave -->, <!-- leave - manual fix -->) or merely has an attribute updated (UNSPECIFIED→NONE), and in two cases the comments were never attached to a removed element at all (<!-- flag --> <!-- add shared-cache-mode … -->). #8353 now reproduces the inline placement the input used:

-  <property name="javax.persistence.sharedCache.mode" value="NONE"/>
-  <!-- change to NONE -->
+  <property name="javax.persistence.sharedCache.mode" value="NONE"/><!-- change to NONE -->

These expectations were asserting the old reflow, so they are updated to match the source. No behavioural change is warranted — the recipe never intended to relocate these comments.

Verification

./gradlew test green locally against the post-#8355 snapshot.