[huhumeng2]

Dropping a Session::BatchGuard currently calls z_session_drop, which explicitly closes the shared session even though the guard only cloned its reference. A second batch can therefore start successfully but its publisher subsequently fails with Z_ERR_SESSION_CLOSED.

Release the guard's reference with _z_session_rc_drop after stopping the batch. Also implement ownership-transferring move operations: the default moves copy the C handle without invalidating the source, and move assignment overwrites the destination's previous reference without stopping its batch.

Add a router-free regression test using isolated loopback TCP session pairs. It covers repeated batches with explicit flush and destructor flush, publishing and declaring a publisher after a batch, move construction, move assignment over an active guard, self-move assignment, and a failed nested start leaving the original batch active.

Related: eclipse-zenoh/zenoh-pico#1315. This addresses the C++ guard lifetime and repeated-batch failure; the separately reported declaration failure during an active batch is not covered by this change.

Validation on Ubuntu 22.04, GCC 11, zenoh-pico d9b4eea486ea90bbeb61dd59cace7bac54a18562:

• New regression test fails on unchanged zenoh-cpp 2fafc38700d256335c1b4f76b3212c42cc70b8c1: assert(!session.is_closed()) after the first batch.
• All zenoh-pico C++ test targets build; the new test and six existing non-router tests pass (7/7).
• New regression passes 20 consecutive runs.
• AddressSanitizer with leak detection passes, with both zenoh-pico and the C++ regression instrumented.
• Strict warning target and clang-format checks pass.
• A separate UBSan attempt stops in pre-existing zenoh-pico protocol/iobuf.h:91 code (null pointer passed as argument 2); UBSan is not claimed clean.

The contributor ECA check has passed.