With the current user transaction sticky tpc retrieval method, a bottleneck can be formed when invocation
returns are too big. To avoid such bottleneck, we'll separate the transaction sticky logic in proxy into two
separate interceptors, one for EJBs, one for ClientUserTransactionService's clustered proxy.

By doing this, we're able to provide a cleaner solution, separating this logic from the proxy, and making it
even more optional, reducing the risks of affecting normal (non user transaction sticky) invocations.

Another interesting idea for the future would be to use a synchronised version of a weak IdentiyHashMap.