core: ExtensionTable drops its filter and best_effort_filter in both proto directions #1382

RelProtoConverter.visit(ExtensionTable) (core/src/main/java/io/substrait/relation/RelProtoConverter.java:424) never calls setFilter or setBestEffortFilter, and ProtoRelConverter.newExtensionTable (core/src/main/java/io/substrait/relation/ProtoRelConverter.java:668) never reads them. The other three read types (NamedScan, LocalFiles, VirtualTableScan) carry both filters in both directions, and the POJO models them on AbstractReadRel.

A read whose mandatory filter must remove every row therefore comes back with no filter, and a consumer returns every row with no error:

ExtensionTable rel =
    ExtensionTable.from(detail)
        .initialSchema(NamedStruct.of(List.of("id"), R.struct(R.I32)))
        .filter(sb.bool(false))
        .bestEffortFilter(sb.bool(false))
        .build();
relProtoConverter.toProto(rel).getRead().hasFilter();          // false
relProtoConverter.toProto(rel).getRead().hasBestEffortFilter(); // false

Decoding a ReadRel{extension_table, filter, best_effort_filter} gives getFilter() == Optional.empty() and getBestEffortFilter() == Optional.empty().

The RelSamples ExtensionTable sample carries no filter, so no round-trip test catches it. #1278 is the copy-on-write side of the same field.

Measured on main at ca8315b. Found while reviewing #1260.