[OPENJPA-2955] Do not remove schema generation keys from the caller's properties map

[rzo1]

extractSchemaGenObjects now records which of the four script keys held a Writer/Reader and strips them from a copy, so the map handed to fromProperties is left untouched — see https://issues.apache.org/jira/browse/OPENJPA-2955. The copy is only made when one of the keys is actually an object, so the common all-strings case allocates nothing.

Both halves of the report reproduce. Map.of(...) really does throw UnsupportedOperationException, and there is an in-tree victim of the mutation as well: DistributedJDBCConfigurationImpl.fromProperties calls super.fromProperties(original) and then keeps using original to build every slice, so each slice silently lost its script Writer/Reader.

New TestSchemaGenObjectProperties covers capture, caller-map immutability, an unmodifiable input map and the plain string path. Three of its four tests fail against master (2 failures plus the UnsupportedOperationException); all four pass here, as do TestSchemaGenDrop, TestSchemaGenAnnotations, TestSchemaGenerationScripts and TestSchemaGenerationProperties (34 tests). The test lives in openjpa-jdbc rather than openjpa-kernel because constructing an OpenJPAConfigurationImpl needs a ProductDerivation on the classpath, and openjpa-kernel has none in its own test scope.