Race conditions when accessing data

[Yadunund]

The current implementation suffers from data races between threads that access the same data, eg. rmw_publisher_data_t. There are essentially two classes of races:

1. One thread destructing the data object and another accessing is modifying it.
2. Two or more threads are modifying the data object.

Proposal to address the issue

• Update all data classes to rely on getters and setters to access members while locking mutexes.
• The type erased data fields in rmw types, eg rmw_publisher_t->data will store the raw pointer to the RMW context. The assumption here is that the context will always be valid if an RMW entity is valid
• The rmw_context_impl_s class will store sets/vectors of rmw_x_data_t objects of the various RMW entities (publishers, subs, clients, servers, etc). Two or more threads will lock the same mutex in the rmw_context_impl_s while accessing/modifying these objects.