[Ada/pushed] Small improvements in ada_evaluate_subexp

[Joel Brobecker <brobecker at adacore dot com>]

Hello, 

I just pushed the following patch series which makes small adjustements
to the handling of OP_VAR_VALUE opcodes in ada_evaluate_subexp.

The motivation for the patch series is to try to avoid returning
"value_zero" objects when in EVAL_AVOID_SIDE_EFFECTS mode, which
is preparation work for better handling dynamic types where the
dynamic properties are expressed using standard DWARF constructs
rather than via GNAT encodings. This will help us out later on,
because some of those dynamic properties rely on having access
to the actual object in order to be resolved (Eg: expressions using
the DW_OP_push_object_address operation). In this case, value_zero
would not allow us to perform that resolution.

As it happens, it better matches what the a standard evaluation does
for OP_VAR_VALUE as well, and the reasons documented via a comment
in evaluate_subexp_standard also apply to the Ada case as well.

The entire series has been tested on x86_64-linux. Pushed.