Block hashes on HSET that don't match the schema of all indices (>=v2)

[K-Jo]

As a module configuration and on an index level.
SCHEMA_MISMATCH_POLICY

Options:

• SKIP current behaviour
• PARTIAL index fields that do match
• BLOCK error on write (HSET)

Note, this could mean we would need a command hook which will introduce a performance degradation

[mikeblum]

This would help with imperfect data tied to each hash - otherwise i need bespoke indexes or extra logic around what can be added to the index in terms of document to get a 💯 % index of every document.

[thotakst]

Partial indexing would be really helpful as finding the exact field that is failing is complex for very large sets of data.

[ashtul]

Related to Partial index 2.0 (#1462)

[oshadmi]

There could be another option for SCHEMA_MISMATCH_POLICY, such as a CONVERT option, which will allow to define conversions for specific mismatches, e.g., allowing one or more fields to accept, let's say a numeric value, by converting it to a string, etc. (whether there are built-in conversions, lua-scripts, or functions)

[neeleshs]

This is a super important feature. Without at least an option to BLOCK error on write (HSET), the clients have no way of knowing indexing this particular doc has failed. hash_indexing_failure counts are unreliable and dont tell which keys failed and thus no way to recover, other than fully rebuilding the index. Productionalizing Redisearch with silent indexing failures is next to impossible