Fix JSON.MSET duplicate-key crash

[roshkhatri]

Fix for: #106

A JSON.MSET that references the same key more than once could crash the server
(SIGABRT). Triples are validated up front and then applied in place under
ValkeyModule_Assert(rc == JSONUTIL_SUCCESS). For a duplicated key, an earlier op changes
the document's shape, so a later op (validated against the original shape) fails at apply
time and the assertion aborts the process.

This removes the apply-phase assertions and applies ops on a best-effort basis. Each key
is opened fresh at apply time, so every op sees the current value, including the effect of
an earlier op on the same key. An op that cannot apply against the current document is
skipped and the remaining ops are still applied; the command returns OK. Up-front
validation is unchanged, so malformed commands are still rejected atomically with an error
before any write.

This also makes JSON.MSET behavior compatible with RedisJSON.

Tests: added cases for a duplicated key whose later op conflicts with the first (returns
OK, server alive, first op committed), a three-op case where a middle op is skipped and
the others applied, and a case where the later op is valid against the first op's result
so both apply. Existing MSET and validation tests pass unchanged; the full integration
suite passes across all server versions including the ASAN build.

[zackcam]

Looks good to me, should we put a note in the docs that the behaviour now will silently fail on a bad duplicate key but still return ok?