[Bug] Pipelining does not preserve positional integrity of replies

[cryo2010]

Summary

flushPipeline and exec return seq[string]:

https://github.com/nim-lang/redis/blob/a0f8216/src/redis.nim#L430
https://github.com/nim-lang/redis/blob/a0f8216/src/redis.nim#L1267

Flattening every reply to strings destroys information that pipeline
callers need:

1. No 1:1 mapping. Status acknowledgments are filtered out and an
   empty array reply contributes zero strings, so N commands can
   produce fewer than N entries. Callers match results to commands by
   position, so any missing slot silently attributes every later result
   to the wrong command:

   r.multi()
   discard r.lRange("cart:items", 0, -1)    # empty list
   discard r.get("cart:owner")              # "alice"
   let res = r.exec()
   # res == @["alice"]: one entry for two commands, and it looks like
   # the answer to the lRange

   Pipelines are dominated by loop-generated batches (N runtime items,
   often several commands per item consumed in strides), where
   positional integrity is the entire contract.

2. Type ambiguity. An integer reply of -1, a nil reply, and the
   literal strings "-1" and "" are indistinguishable after
   flattening, and redisNil ("\0\0") is an in-band sentinel that
   collides with real data.

3. Structure loss. A reply that is an array of arrays arrives as
   one flat list with the boundaries erased.

These are properties of the seq[string] return type itself and cannot
be fixed behind the current signatures.