This PR adds a feature called serve-original-ttl which, if enabled, will make Unbound return the original TTL for records as received from upstream. This feature may be useful if Unbound serves as a front-end to, e.g., a hidden authoritative name server.

Notes on the PR:

• The caching logic was not changed, the TTL values in packed_rrset_data structures are treated the same as before. Instead, the new code stores an additional value in this structure call ttl_add which records how much has been added to the TTLs in the structure, and can be subtracted from these absolute TTLs instead of the current time when encoding responses.
• The logic in packed_rrset_encode in msgencode.c was changed slightly to make the time adjustment explicit. The original code sometimes modified the value of the function's input parameter timenow, setting it to 0. With the introduction of the new feature, this would have led to messy conditionals when adjusting the TTLs during encoding.
• Based on feedback from @wcawijngaards on an earlier version of this code the logic when decrementing TTLs in cachedb/cachedb.c was also changed such that the semantics of the decrement should be preserved while the original TTL will still be returned if this feature is enabled.
• The manual page explicitly specifies that enabling this feature means Unbound will ignore cache-min-ttl and cache-max-ttl.
• When allocating space for struct packed_rrset_data the code was changed in several places to explicitly initialise the newly allocated memory to zero, using either regional_alloc_zero or calloc(1, <size>).