fix(string_family): Add RemoveDoc for the SET command

[BagritsevichStepan]

fixes #5215

The problem was that the StringFamily::Set command could change the key's type. If the previous type was HSET or JSON, the key was indexed. When StringFamily::Set is called on such a key, it changes the type from HSET/JSON to STRING, but we forgot to remove the key from the indexes.

This PR adds a call to RemoveDoc inside StringFamily::Set to properly clean up the old index entry.

A snippet that reproduces the issue has been added to the tests, along with explanatory comments:

127.0.0.1:6379> ft.create index on hash schema field TEXT
OK
127.0.0.1:6379> hset k1 field value
(integer) 1
127.0.0.1:6379> set k1 anothervalue
OK
127.0.0.1:6379> rename k1 anotherkey
OK
127.0.0.1:6379> hset k1 field value
Error: Server closed the connection
(1.29s)

[vyavdoshenko]

@BagritsevichStepan
Could you please add a more descriptive description of the root cause and how it was fixed?

[BagritsevichStepan]

@BagritsevichStepan Could you please add a more descriptive description of the root cause and how it was fixed?

Done

[vyavdoshenko]

LGTM

@romange
Could you please take a look at this PR as well?

I see the potential issue. In my opinion, this issue can be bigger than this specific issue. We have to track the search index in all possible places, briefly speaking, all mutable operations. I am afraid that the same issue can cause the same or similar crash during the change of a different pair of types.
Personally, I don't know how to generalize the approach and avoid similar issues in the future. Maybe I am paranoid, correct me if I am wrong.

[BagritsevichStepan]

@vyavdoshenko I spoke about it today and we already have an issue for this #5014 . I have a solution

[romange]

i do not like this unconditional approach. please check before if prime_value.ObjType() is indexable (i.e. either hset or json). You can introduce a one-liner utility function for better readability (add it to family_utils.h).

[BagritsevichStepan]

I know, we have this problem in several places (check for example rename command). Let's do it in another PR, I will fix other commands as well.

[romange]

ok, please do not forget

[romange]

most of our write commands are not blind writes so the issue is not wide to begin with.

[BagritsevichStepan]

I know for example zunionstore command, that also changes the key type

[BagritsevichStepan]

And there is the same bug in the OpAdd in set_family.cc if overwrite is true