Fix leader-switch race in v_chunk_id IPC publish within create_shard

[Copilot]

The v_chunk_id publish step in create_shard() used run_on_pg_leader(), which has a TOCTOU gap: if the leader changes between get_stats() and the lambda, nobody executes it and auxiliary_uint64_id stays at UINT64_MAX, causing the subsequent v_chunk_id equality assertions to flap.

Change

• Replace run_on_pg_leader() with run_on_pg_leader_with_retry() for the v_chunk_id IPC publish step, using auxiliary_uint64_id != UINT64_MAX as the done_check — consistent with how create_shard itself was already fixed.

// Before
run_on_pg_leader(pg_id, [&]() {
    auto v_chunkID = _obj_inst->get_shard_v_chunk_id(shard_id);
    RELEASE_ASSERT(v_chunkID.has_value(), "failed to get shard v_chunk_id");
    g_helper->set_auxiliary_uint64_id(v_chunkID.value());
});

// After
run_on_pg_leader_with_retry(pg_id,
    [&] { return g_helper->get_auxiliary_uint64_id() != UINT64_MAX; },
    [&]() -> bool {
        auto v_chunkID = _obj_inst->get_shard_v_chunk_id(shard_id);
        RELEASE_ASSERT(v_chunkID.has_value(), "failed to get shard v_chunk_id");
        g_helper->set_auxiliary_uint64_id(v_chunkID.value());
        return true;
    });