Hide data representation inside RDKit::Dict (#9113)

* Remove Dict::getData() for a strict abstraction boundary

Replace direct access to Dict's internal std::vector<Pair> with
encapsulated methods: size(), empty(), const iteration via
begin()/end(), appendPair(), markNonPOD(), and getRawVal().

This enables future changes to Dict's internal representation
without breaking callers.

Ref: rdkit/rdkit#9112

* Harden Dict::appendPair to take a populated Pair by move

appendPair(Pair&&) now auto-detects non-POD status via
RDValue::needsCleanup(), eliminating markNonPOD() and the
risk of dangling references or uninitialized entries.

needsCleanup() is placed next to destroy() on RDValue to
keep the POD/non-POD distinction in one place.

* Remove vestigial dictHasNonPOD param from streamReadProp

Both callers ignored the output. Non-POD detection is now handled
by Dict::appendPair via RDValue::needsCleanup().

* unbork java build

* Address PR review: bulk append, rename getRawVal, add custom data test

- Add Dict::append(vector<Pair>&&) for bulk insertion with reserve
- Use bulk append in streamReadProps to restore pre-allocation
- Rename getRawVal -> getRDValue per reviewer preference
- Add test verifying custom AnyTag data is destroyed through Dict lifecycle

* heed self-review

* don't manually implement vec.insert

* Add test: ExplicitBitVect round-trip through Dict serialization

Exercises the full streamWriteProps/streamReadProps path with an
ExplicitBitVect in an RDProps Dict, confirming the custom handler
is invoked and no memory is leaked (verified under valgrind).

* in anyTag test, assert destructors ran a specific number of times.

---------

Co-authored-by: bddap (Coding Agent) <andrew+bot@dirksen.com>

Code/JavaWrappers/Dict.i

@@ -40,6 +40,11 @@
 
 %ignore RDKit::Dict::Pair;
 %ignore RDKit::PairHolder;
+%ignore RDKit::Dict::extend;
+%ignore RDKit::Dict::getRDValue;
+%ignore RDKit::Dict::begin;
+%ignore RDKit::Dict::end;
+%ignore RDKit::Dict::const_iterator;
 %include <RDGeneral/Dict.h>