Separate id used for internal trace map from transaction id used for context propagation

[Xylus]

A single id was used for both transactionId for context propagation, and trace map cache (as key).
This approach however does not work well for cases when a transaction is continued. For a continued transaction, using the transactionId as the map key may conflict with transactionId issued for new transactions.

Separating id used for trace map key (at least for continued transactions) would fix the problem.
This should be done using the least amount of atomic counters that covers id generation for 4 cases:

1. New Sampled Trace
2. New Unsampled Trace
3. Continued Sampled Trace
4. Continued Unsampled Trace

[Xylus]

Generate id differently (currently divides the Long space into 4)
Better to leave some space for future use.