[OPENJPA-2957] Resolve embeddables through the metadata repository instead of annotation names

[rzo1]

Replaces the simple-name annotation scan in AbstractExpressionBuilder.hasEmbeddableAnnotation with a repository lookup, as suggested in https://issues.apache.org/jira/browse/OPENJPA-2957. ClassMetaData.isEmbeddable() is set by both the annotation and the XML parser, so it also covers embeddables declared only in orm.xml and no longer matches any annotation that happens to be called Embeddable.

Note that fmd.getEmbeddedMetaData() cannot be used directly here: the guard is only reached when that call already returned null, so the check has to ask about the declared type instead. This is a robustness fix rather than a behavioural one — I could not construct a case where the old check produced a wrong answer, since the branch is only reachable for an @EmbeddedId field whose embedded metadata is absent. openjpa.persistence.embed.** (89 tests, including the XML-mapped ones) and the TestDerivedIdEx2b / TestUnenhancedDerivedIdEx2b regression tests for the original 2.4.1.3 ex2b fix are green.