isthmus-cli: SqlLibraryOperators is not registered for reflection, so every library-operator mapping is unreachable in the native image

[nielspardon]

SqlLibraryOperatorTableFactory discovers operators by reflecting over the public fields of SqlLibraryOperators, but isthmus-cli's RegisterAtRuntime registers SqlStdOperatorTable.class and not SqlLibraryOperators.class. In the native image that field scan therefore yields nothing, SubstraitOperatorTable's LIBRARY_OPERATOR_TABLE ends up empty, and every mapping backed by a library operator is unreachable — while the JVM test suite stays green, because it never runs against the image.

Verified against a nativeCompile -Pquick-build-native build of main, with --create 'CREATE TABLE strings (c16 CHAR(16), vc32 VARCHAR(32), vc VARCHAR)':

select reverse(vc32) from strings   -> Error: No match found for function signature REVERSE(<CHARACTER>)
select sinh(1.0)     from strings   -> Error: No match found for function signature SINH(<NUMERIC>)
select least(i, i)   from t         -> Error: No match found for function signature LEAST(...)
select starts_with(vc32, 'a') ...   -> Error: No match found for function signature STARTS_WITH(...)
select lpad(vc32, 4) from strings   -> Error: No match found for function signature LPAD(...)
select log2(d)       from t         -> Error: No match found for function signature LOG2(<NUMERIC>)

select upper(vc32)   from strings   -> "name": "upper:vchar"        (SqlStdOperatorTable — works)
select initcap(vc32) from strings   -> "name": "initcap:vchar"      (SqlStdOperatorTable — works)

So the split is exactly "is the Calcite operator declared in SqlStdOperatorTable or in SqlLibraryOperators", and roughly a dozen SCALAR_SIGS entries fall on the wrong side of it: the hyperbolic family, least/greatest, log2, factorial, concat, starts_with/ends_with/contains, left/right, lpad/rpad, and now reverse.

Adding SqlLibraryOperators.class to the register(...) block in RegisterAtRuntime was verified sufficient — with it, reverse:vchar, reverse:fchar and sinh:fp64 all resolve in the rebuilt image.

The reason this survived is that isthmus-cli's native smoke tests only cover TPC-H and arithmetic, so no case exercises a library operator. A smoke case for one (sinh or starts_with) belongs with the fix, otherwise the same gap reopens silently.

Found while reviewing #1251, which adds the reverse mapping and is the newest entry to land on the broken side.