get rid of sds arguments in SortedMap

[romange]

(good for starter tasks)

SortedMap still has some methods with sds because zset_family code uses sds and that in turn due to reliance on zzl utlilites like zzlFind

1. We should reimplement zzl utilities using string_view instead of sds, (only those that we need of course),
   and use where they are needed in zset_family, similarly to how we implemented LpFind vs lpFind
2. We should get rid of SdsWrap calls in zset_family and pass string_view directly.
3. We should get rid of functions accepting sds arguments inside SortedMap.

An example PR that does such cleanup: #4638

[abhijat]

@romange I have a WIP pr to do this cleanup #4738, there is an issue I see there which is seen when inserting empty strings in sorted map:

1. during argument parsing empty strings are passed around as default initialized string_view - these return nullptr when data() is called on them
2. when storing field with score we do an lpAppend call which inserts before end of list pack. We pass the result of calling data() on default initialized string view here

   dragonfly/src/core/sorted_map.cc

   Line 193 in 24b3723

   return zzlInsertAt(zl, NULL, (unsigned char*)ele.data(), ele.size(), score);

3. When a null pointer is inserted to list pack, it is treated as a replacement operation which is incorrect in our case and also triggers assertion failures

   dragonfly/src/redis/listpack.c

   Line 836 in a3cabd1

   int del_ele = (elestr == NULL && eleint == NULL);

To avoid this in the PR, we handle the corner case by detecting a null pointing string view before lpAppend and convert it to string_view{""}.

This seems to work fine for this use case, but just wanted to call this out in case maybe there are other ways to approach this or even keeping WrapSds around for this specific code path.

[romange]

yes, redis code does not handle well nulls with zero lengths. we had to handle it explicitly several times.