reset next_size/start_size when calling purge_memory

[mark05417]

int main() {
    boost::pool<mark_new_delete> memory_pool(sizeof(int));
    
    for(int i=0;i<40;i++){
        // use malloc to get some memory from pool
        int* data = static_cast<int*>(memory_pool.malloc());
        int* data2 = static_cast<int*>(memory_pool.malloc());

        // release all the memory to the system instead of calling memory_pool.free(data)
        memory_pool.purge_memory();
    }
    return 0;
}

Consider the above code, I think it is a normal scenario that users may use, it has the performance issue.

Althought the more correct usage might be call memory_pool.free() to free those memory (data1, data2) one by one instead of calling purge_memory() at each iteration/rounds.

The current doubling algorithm doesn't reset the start_size and the next_size. If calling memory_pool.purge_memory(), the pool releases memory to the system. And the next time memory_pool.malloc() is called, the pool asks for the doubled memory from system, which is not expected.