`ekaf_picker:pick_sync` does not consider custom strategy

`ekaf_lib:common_sync/4` uses `ekaf:pick(Topic)` -> `pg2:get_closest_pid(Topic)` which does not take `custom` strategies into account. This makes `ekaf:produce_sync*` functions work differently for a binary key vs a `{binkey, binval}` tuple:

``` erlang
158> ekaf:produce_sync_batched(<<"hai">>,<<"value">>).
{buffered,0,1}
159> ekaf:produce_sync_batched(<<"hai">>,{<<>>,<<"value">>}).
>>>>>>>>>>>>>>>>>>> {<<"hai">>,
{<<>>,<<"value">>},
[<0.4644.1>,<0.5064.1>,<0.8221.1>]}
[{buffered,0,1}]
```

The async variants already work as expected in all cases. The partitioner might want to do it's own default thing for all `ekaf:produce_*` calls.

Reproduce by setting `ekaf_callback_custom_partition_picker` to:

``` erlang
partition_picker(Topic, Data, State) ->
io:format(">>>>>>>>>>>>>>>>>>> ~p~n",[{Topic, Data, State}]),
{partition, 0}
.
```