JMSPublishFilters save memory allocations and CPU cycles by caching null values for non-existant properties

[eolivelli]

There are two places in which we use Map.computeIfAbsent in the JMSPublishFilters code path.
The problem with computeIfAbsent is that null values are not cached and so the broker ends up in scanning the whole list of properties in case of a selector that refers to a properties that is not set on the message.

Motification:

• cache a dummy sentinel value in order to not redo the search every time
• eagerly pre-cache as a Map all the properties of the Message in case the Message is to be processed by more than 1 Subcription