Feature/alias as index attribute

[rileydes-improving]

Summary

Adds FT.ALIASADD, FT.ALIASDEL, and FT.ALIASUPDATE with RediSearch-compatible behavior. Also adds FT.ALIASLIST (not present in RediSearch) for listing all active aliases. Aliases live as a repeated field on the owning IndexSchema proto, persist in RDB as part of the index, and propagate cluster-wide through the existing MetadataManager / FT.INTERNAL_UPDATE path. No separate metadata type needed.

Changes

Proto & IndexSchema

• Added repeated string aliases = 15 to IndexSchema in index_schema.proto
• Extended IndexSchema class with GetAliases() / SetAliases() accessors and in-memory alias list

SchemaManager — Alias CRUD & Resolution

• Added db_to_aliases_ (flat_hash_map<uint32_t, flat_hash_map<string, string>>) as a forward alias map (alias → index name) per database
• Implemented AddAlias, RemoveAlias, UpdateAlias, and GetAllAliases
• In coordinator mode: each alias operation fetches the owning index's proto from MetadataManager, mutates the aliases repeated field, normalizes defaults, and re-commits via CreateEntry, which triggers OnMetadataCallback cluster-wide
• In standalone mode: directly mutates the in-memory alias map and IndexSchema's alias list under the mutex
• GetIndexSchema falls through to alias resolution when direct name lookup misses
• RemoveIndexSchemaInternal purges all aliases for a dropped index via EraseAliasesForIndex
• OnMetadataCallback uses MessageDifferencer (ignoring aliases and stats fields) to detect alias-only changes: when structural fields are unchanged, it calls RebuildAliasMapsForIndex and SetAliases without tearing down the index
• ComputeFingerprint clears the aliases field before hashing so alias mutations don't destabilize search consistency checks
• OnFlushDBEnded in coordinator mode fetches the authoritative proto from MetadataManager (including aliases) when recreating indexes after FLUSHDB
• OnSwapDB swaps alias maps in lockstep with index maps

FT.INTERNAL_UPDATE, Type Routing (Requested by Allen)

• Extended FT.INTERNAL_UPDATE to accept optional keyword/count/args-style parameters after the required 4 arguments. Currently recognizes TYPE <1> <type_name> (defaults to kSchemaManagerMetadataTypeName for AOF backward compatibility).
• ReconcileMetadata now skips replication for types not registered locally, preventing errors on nodes that don't recognize a metadata type.

Commands

• FTAliasAddCmd, FTAliasDelCmd, FTAliasUpdateCmd in ft_alias_add.cc, ft_alias_del.cc, ft_alias_update.cc, registered in module_loader.cc with @search @write @fast ACL permissions
• FTAliasListCmd in ft_alias_list.cc, returns all alias->index pairs for the selected database
• In cluster (CME) mode, ALIASADD/ALIASUPDATE perform a consistency-check fanout (AliasExistsConsistencyCheckFanoutOperation). ALIASDEL uses AliasRemovedConsistencyCheckFanoutOperation to wait until the change is visible on all primaries.
• MULTI/EXEC and Lua are rejected in CME mode to avoid local-only mutations without fanout
• ft_dropindex.cc resolves an alias to its real index name before calling RemoveIndexSchema
• ft_aggregate.cc uses the alias field when constructing its response

FT.INFO

• ft_info_parser.cc calls GetAliases() on the IndexSchema and reports a sorted aliases array in FT.INFO output

Testing

• Unit tests in ft_alias_commands_test.cc: argument count errors, happy paths, duplicate rejection, alias-to-alias rejection, self-referential alias, and ReplicateVerbatim behavior in both coordinator modes
• Unit tests in schema_manager_alias_property_test.cc: property-based tests for add/remove/update round-trips, cross-database isolation, upsert semantics, FLUSHDB/SWAPDB behavior, alias-only metadata callback, and fingerprint stability
• Unit tests in ft_info_test.cc: sorted aliases array and querying via alias still reports the real index name
• Integration tests in test_ft_alias.py: all four commands, DROPINDEX cleanup, FT.SEARCH/FT.AGGREGATE via alias, FLUSHDB, SWAPDB, RDB persistence, wrong argument errors, and name-collision edge cases
• Integration tests in test_ft_alias_cluster.py: ALIASADD, ALIASDEL, ALIASUPDATE, ALIASLIST, and DROPINDEX tombstoning all propagate to every cluster primary
• Compatibility: alias-answers.pickle.gz added to the parametrized run in compatibility_test.py. Alias management commands are replayed inline to reconstruct alias state before dependent search/aggregate queries.

Architecture Difference vs. PR valkey-io#955

PR valkey-io#955 introduced a separate metadata type (vs_alias) with its own MetadataManager registration, OnAliasMetadataCallback, and dedicated RDB_SECTION_ALIAS_MAP section. This PR stores aliases directly as a repeated field on IndexSchema, treating alias mutations as proto updates through the existing vs_index_schema metadata pathway. That removes the need for a separate metadata type, a separate RDB section, and the ordering constraint (indexes must load before aliases). Alias changes propagate via the same CreateEntry call used for index creation. OnMetadataCallback detects alias-only changes via MessageDifferencer and applies them without index teardown.

Example

FT.CREATE my_idx ON HASH PREFIX 1 doc: SCHEMA category TAG
HSET doc:1 category books
FT.ALIASADD my_alias my_idx
FT.SEARCH my_alias "@category:{books}"

Returns 1 result (doc:1) via the alias without any change to the underlying index.

[AlexFilipImproving]

The code looks good, but for consistency with the rest of the codebase, can you combine the alias commands into a single ft_alias.cc file, like the other commands? That would be add, del, list and update. You can put the ft_alias_consistency.h file at the top so you don't have to include a separate .h and it will make it easier to search in the future.

[rileydes-improving]

The code looks good, but for consistency with the rest of the codebase, can you combine the alias commands into a single ft_alias.cc file, like the other commands? That would be add, del, list and update. You can put the ft_alias_consistency.h file at the top so you don't have to include a separate .h and it will make it easier to search in the future.

Made the change, all in one file now and centralized some of the functions.