Answer the MobilityDB operators soundly from a multi-entry key

[estebanzimanyi]

A key of a multi-entry index covers one entry of the indexed value, while the consistent methods of MobilityDB answer for a key that covers the whole value. The operator classes for the temporal, the temporal number and the temporal point types read those methods, and two of their answers do not hold for an entry: contains and same ask the key to hold the query, which no single entry need do even when the value holds it, and the position strategies are exact for a key covering the whole value, so MobilityDB reports that they need no recheck. The multi-entry methods state both corrections, answering contains and same with overlaps and rechecking every strategy, and delegating the rest to MobilityDB, which keeps one implementation of the box semantics. The topological and the position operators of MobilityDB answer on the bounding box of a value; the entries of a value cover the value but not its bounding box, so a query that meets the box where no entry lies has no entry that can find it, and the bounding box is itself a key of the value. Measured on tbl_tint with num_boxes=3, contains reads 35 rows against 105 sequential; tbl_tgeompoint rows 33 and 31 overlap as boxes while no entry of one meets the box of the other, which reads 151 rows against 152. The whole contrib suite, 28 tests, agrees with the sequential scan against a MobilityDB that carries the split-box bound inclusivity fix.