Residue from the @service phase split: unpinned Scalar trigger, dropped Table filter, stale catalog test names

Three small things the @service coordinate-precedence phase split left behind. Filed at its Done gate as follow-ups because none of them breaks a contract; each is a place where the split’s new shape is not yet fully pinned or fully honest. Independent of each other, so this can ship in pieces or be split.

The RecordParent classify arm’s second trigger is half-pinned. ServiceDirectiveResolver.classifySourcesCoordinate’s record-parent arm rejects on either of two triggers: a SOURCES candidate, or a `Result / Scalar resolved return type with no candidate in sight. The second trigger is the over-fire guard that keeps a Sources-less record-parent @service from silently becoming legal a phase early, and only its Result half is pinned (ServiceCoordinatePrecedenceTest.recordParent_noBatchParameterWithRecordReturn_isStillRejected, whose fixture returns a record-backed type). A Scalar-return sibling with no candidate would close it. One fixture, same mould as its neighbours.

inferBindingsByType’s decoded overload dropped the `org.jooq.Table eligibility filter. The reflection-based overload (still live on the @condition path) skips Table<?> parameters when choosing inference candidates; the decoded overload the @service claim reduction calls does not, because ServiceCatalog.DecodedParam carries an isDslContext flag and no is-a-Table flag. Harmless today: a Table<?> parameter on a @service method is rejected either way, and only the message differs (an inferred binding’s extraction rejection instead of the argument-name mismatch). Worth closing anyway, because the two overloads now read as equivalent forms of one rule and are not. Either add the flag to DecodedParam and filter on it, or state in the decoded overload’s javadoc why the axis is absent there.

ServiceCatalogTest’s method names still say `reflectServiceMethod. 41 test methods carry the reflectServiceMethod_ prefix naming an entry point that no longer exists; they drive the class’s own reflect(...) helper, which composes decodeServiceMethod + reduceClaims + bindServiceMethod. The javadoc on the helper explains the composition, so nothing misleads a reader who looks, but the names are a prose surface naming a dead symbol. Deliberately left out of the split’s own commit: 41 renames is a wide mechanical diff with real rebase-conflict cost against concurrent sessions, and it buys clarity rather than correctness. Rename them as a standalone commit when the file is otherwise quiet, splitting the prefix by what each case actually drives (decode_, bind_, or decodeAndBind_).