lib-sieve: plugins: duplicate: Only save duplicate entry when the last script executed successfully.

This prevents prematurely marking duplicates when errors occur during result
execution. Before, the pending duplicate database changes were committed after
each script executed in the sequence. If a later script failed, duplicate tests
would mark duplicate messages prematurely.

Disadvantage is that actions in earlier scripts that are shielded by a duplicate
test can be executed twice erroneously when errors occurred in later scripts in
the previous delivery. Still, unexpected duplicate action execution is better
than losing mail due to erroneous duplicate detections.

Fixing this properly requires extensively restructuring result execution.

This indicates whether the last script in the sequence is being executed.

Before, it would only terminate the test when it failed explicitly using the
test_fail command.

The actions contained in the result may keep references to the overriden mail
struct, so these need to be preserved.

The actions contained in the result may keep references to the overriden mail
struct, so these need to be preserved.

This holds a reference to the initial result, causing it to linger after
test_result_reset, thereby also holding a reference to the message context,
preventing it from being cleared. This in turn, can keep references to modified
message versions that cause the main message object to remain referenced. This
causes a panic in specific cases upon test_result_reset:

Panic: Input stream data unexpectedly has references

Caused a panic when input was not valid. This is only relevant to the test suite
and doesn't occur anywhere else (apart from the previous similar commit for
lib-sieve-tool).

Panic was:

Panic in file smtp-address.c: line 684 (smtp_address_write): assertion failed: (smtp_char_is_qpair(*p))`

It is immediately referenced in sieve_interpreter_continue(), called from there.

The error string is currently always used and it must be, so there is no need to
allow for the error_r paramerter to be NULL and unused.

The segfault happens when this match type is the last argument of the test
command. This situation is not possible in a valid script; positional arguments
are normally present after that, which would prevent the segfault. A variant of
this bug occurs when the match type also has no argument of its own. In either
case the segfault is caused by referring to absent tag arguments, which causes
a NULL dereference.

It's still not fully correct, but we don't have existing code that supports
parsing it.