This introduces a struct for VT references.
The nvti objects is extended to have a list of the vtref
objects.
Basically the new struct is handled similar to the prefs.

Also, the xref element of nvti is removed and the API
uses the new list of vtrefs to mimic the previous behaviour.
Actually this is transitional until all modules will use
the new generic references.

Also improves the xref handling slightly.

This makes them availble to other modules.

This new method allows to add a bulk of comma-separated
references to the NVTI. All references share the same
type and (optional) text. A typical use case is to
add bulk of CVEs.

These exception are from old OTP times and the keywords
should not appear anymore.

The code in nvti_add_bid and nvti_add_cve is meanwhile
redundant with the code in nvti_add_refs_from_csv
and thus these methods now simply call the latter.
Eventually they can be eliminated once gvmd does not call
them anymore.

Also occured the need to const'ify vtref_new and
nvti_add_refs_from_csv for doing so.

And use nvti_add_refs_from_csv instead of nvti_set_bid.

This removes nvti_set_cve(), nvti_set_bid(),
nvti_add_cve() and nvti_add_bid().

One thing that needs to be considered when changing this
in modules using this function is to free the returned string.

This enables nvti_ref() to exclude types and to use
either simple CVS or a CSV that includes the type
of each reference.

This allows to remove nvti_set_xref.

This makes the naming consistent with other functions.

For compatibility reason the CSV is using
", " instead of just ",". This is the syntax
used before and other tools using the redis-database
directly rely on this syntax.

This add vref_type(), vtref_id(), nvt_ref() and
nvt_ref_len() to the API.
These functions allow to work with the actual reference
object rather than with the serialized strings.
The names are choosen to be consistent with the prefs
handling.

This reduces the number of code lines slightly for the
g_strsplit loops.
Also adding missing free's for the newly allocated split strings.

Since the element of the struct should not be used by
other than the internal functions, move the struct out of
the header file into the module.

This renaming is for consistent naming scheme.

The function g_malloc0() already does the check.

Removed the "A copy will be created" notes from the doc-string where const
is used anyway.

The split was done for every refrence for the very same string.
Now it is done only once.

Also fix a confusion about renaming function nvti_add_vtref.

It is already done in the for ().