Fix NoGuava* collection recipes crashing on Groovy sources

[timtebeek]

What

org.openrewrite.java.migrate.guava.NoGuavaMapsNewHashMap throws on Groovy sources:

java.lang.IllegalArgumentException: Expected a template that would generate exactly one statement
to replace one statement, but generated 0. Template:
new HashMap<>(__P__.<java.util.Map>/*__p0__*/p())
...
  org.openrewrite.java.migrate.guava.NoGuavaMapsNewHashMap$1.visitMethodInvocation(NoGuavaMapsNewHashMap.java:69)
  org.openrewrite.java.migrate.guava.NoGuavaMapsNewHashMap_1_GroovyVisitor.visitMethodInvocation(...)

Found via the Moderne SaaS flagship run of Java best practices against the "Netflix + Spring" org, where it puts Netflix/metacat into RepositoryRecipeRunError on every run. The trigger is metacat-thrift/src/test/groovy/com/netflix/metacat/thrift/CatalogThriftHiveMetastoreSpec.groovy — a Spock spec containing Maps.newHashMap(ImmutableMap.of(...)), which a sibling recipe in the same composite (NoGuavaImmutableMapOf) has already rewritten to Maps.newHashMap(Map.of(...)) by the time this recipe sees it.

Why

The templates are marked .contextSensitive(). That makes JavaTemplate reconstruct and re-parse the enclosing block as Java before splicing in the replacement. On a Groovy LST that block routinely contains constructs with no Java equivalent — def result = ..., timer.record(_, _) >> {}, registry.createId(_ as String) >> id — so the reconstructed source does not parse, parseBlockStatements returns zero statements, and maybeReplaceStatement throws.

None of these templates reference anything from the surrounding scope: they are self-contained expressions (new HashMap<>(#{any(java.util.Map)})) whose only external dependency is an import they already declare explicitly. Context sensitivity was never needed here, and dropping it also lets the templates be cached.

• I went with the real fix rather than a language guard (cf. openrewrite/rewrite@e9c478e / rewrite-static-analysis#105, which caught RecipeRunException and skipped) — the replacement works correctly on Groovy once the template is context-free, so there is no reason to skip those sources.

Scope

The same unnecessary .contextSensitive() appears across the whole NoGuava* collection-factory family, all with the identical latent crash, so they are fixed together:

• NoGuavaListsNewArrayList, NoGuavaListsNewCopyOnWriteArrayList, NoGuavaListsNewLinkedList
• NoGuavaMapsNewHashMap, NoGuavaMapsNewLinkedHashMap, NoGuavaMapsNewTreeMap
• NoGuavaSetsNewConcurrentHashSet, NoGuavaSetsNewHashSet, NoGuavaSetsNewLinkedHashSet

NoGuavaDirectExecutor is deliberately left alone — its Runnable::run method reference does lose type attribution without context (MemberReference type is missing or malformed), so .contextSensitive() is load-bearing there.

Testing

Added NoGuavaMapsNewHashMapTest#replaceWithNewHashMapWithMapInGroovy, a groovy(...) source spec modelled on the metacat block (Spock when:/then:, mock interactions, multi-line argument) that reproduces the crash before this change. It needs a GroovyParser with Guava on its classpath, and an explicit testImplementation("org.openrewrite:rewrite-groovy") so org.openrewrite.groovy.Assertions is no longer picked up only transitively.

All 218 tests in org.openrewrite.java.migrate.guava pass.