Fix memory leak in ValidateDTD m_dtd

[martinhsv]

This pull request addresses the memory leak listed as 3) in #2469 .

The function call in this line of code allocates heap memory. The second and subsequent calls to this function resulted in the the m_dtd pointer simply being assigned a new value, which meant that the heap memory previously pointed-to was leaked on a per-transaction basis.

m_dtd = xmlParseDTD(NULL, (const xmlChar *)m_resource.c_str());

With this new code, before the new parse-and-assign, we'll check if m_dtd is not null, and if so do the proper cleanup.

(I have addressed the leak in this pull request, but a larger issue we may want to consider for later implementation: Do we really want to be calling xmlParseDTD for every transaction? Granted, if we only want to do this once rather than per-transaction, we would need to consider the use case where the resource content changes over time -- particularly if it is an internet resource. We could decide we don't care, or perhaps adopt a cacheing-and-periodic-update strategy)

[zimmerle]

As mentioned on #2473 I am afraid that the leak of const utilization on the libxml may be an impediment to actually have those optimized on run-time (evaluate) as oppose to configuration time (init). The plan is to have evaluate as const in the operator context. Lets have a chat on #2473 and we get back to this one.

[martinhsv]

Closing this in favour of #2478