feat(database): add Compatible indexes for Mongo and SQL dialects

[ChrisPdgn]

Description

Rewrite of the approved #643 index work from current origin/main (not a cherry-pick/rebase of the stale indexes branch).

Schema indexes are now first-class on MongoDB, PostgreSQL, MySQL, MariaDB, and SQLite. Platform models (authorization + chat) declare CompatibleIndexType Ascending/Descending so the same indexes exist on SQL too.

What this does

• Adds CompatibleIndexType (Ascending / Descending) plus optional unique. Maps to Mongo 1/-1 and SQL BTREE ASC/DESC.
• Index identity is fields in order + unique. Name is a label: adopt the live engine name when that identity already exists, otherwise generate a deterministic cnd_idx_* / cnd_uidx_* name.
• Register (platform and custom-module gRPC) binds declared indexes to live Mongo/SQL names before create/sync(), so upgrades do not throw Mongo 86 or create a second SQL btree.
• Admin extras in _DeclaredSchema survive module re-register. Stale generated names for an identity now bound to a live name are dropped. Leftover live indexes are not dropped on register; use deleteIndexes.
• Admin create / get / delete / import / export persist via _DeclaredSchema. No full schema rebuild/sync() for Admin index create/delete.
• Create/delete persist only what the engine actually applied (including identity-bound skips). Unique-data / options-conflict do not persist. Later failure still persists the applied prefix, publishes, then throws.
• Persist re-reads _DeclaredSchema from primary. SQL same-name conflicts (42P07 / 1061) reuse only when fields+unique match; same name with different fields throws.
• getIndexes uses the live engine as source of truth and overlays declared types by name, then by identity.
• After persist (and after register bind/save), publishSchema notifies replicas. registeredSchemas is snapshotted after bind/save so replica catch-up sends live names. Replicas use instanceSync=true and do not create indexes.
• SQL adapters use originalSchema.collectionName everywhere (no hardcoded cnd_ prefix).
• Admin create is privileged for unique indexes. Import is not privileged and respects schema owner.
• Export/import are Admin-only and paginated (skip/limit). Same-name indexes are skipped on import; create still binds by identity so it cannot duplicate a live index on the same fields.
• validateModelOptions accepts indexes and still allows conduit.readPreference.
• Mongoose dropIndex is awaited. autoIndex is off.
• On schema recover, foreign-dialect leftover types are warned and skipped. Admin-bound invalid types throw.

SQL scalar Relation indexes
Sequelize stores a scalar Relation such as ChatMessage room as roomId. Indexes now follow that mapping instead of skipping every Relation:

• Declared field room ↔ engine column roomId for SQL bind / create / get. Live roomId compounds adopt the declared room identity.
• Index names are generated from declared fields only (room, not roomId), so Mongo and SQL keep the same cnd_idx_* labels.
• Authz String *Id fields (resourceId, subjectId, …) are not rewritten. Only actual scalar Relation fields get the Id suffix.
• Array / extracted Relations stay unindexed on the parent table. ChatRoom participants compounds are still skipped.
• MySQL JSON Compatible btree stays rejected (e.g. ObjectIndex inheritanceTree).
• Field-level scalar Relation indexes are lifted to modelOptions.indexes before extractRelations. Field-level array Relations are still skipped.
• Admin create/import canonicalize inbound roomId → room when room is a scalar Relation. getIndexes / export return declared names.
• Create remains primary-only. instanceSync is unchanged (replicas do not create indexes).

Related SQL register fixes on this branch

• Chat module register is Promise.all again.
• SQL ChatRoom / ChatParticipantsLog register no longer deadlocks waiting for the related model to sync() first.

Intentionally not changed / not ported

• getDatabaseType() still returns 'PostgreSQL' (no 'postgres' rename).
• No Client API, no BullMQ, no masterkey.
• Does not conflate with ReBAC ActorIndex / ObjectIndex documents — only their schema index declarations changed to Compatible.
• Did not port old feat(database)!: add index support for all sql dialects & index refactoring #643 bugs: case 'mysql' || 'mariadb', metadata-only getIndexes, createSchemaFromAdapter rebuild for indexes, unbounded export, required index names, import always treating unique as database-owned, ReverseMongoIndexTypeMap display-name round-trip.
• Out of scope: cross-dialect export/import, HASH-on-MySQL as a product goal, renaming leftover live indexes, restoring field-level fields.*.index from module source after Admin deleted that field index, indexing array/extracted Relations on the parent table, MySQL JSON Compatible btree.

Fixes #643 (implementation rewrite; this PR does not close or merge #643 by itself).

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

How Has This Been Tested?

Unit tests in modules/database/src/__tests__/indexes/ (bind, unique-collision vs name-exists, extras union, partial persist, overlay-by-identity, primary persist read, SQL same-name/different-fields, publish, declared room ↔ engine roomId, Authz String *Id safety, ChatRoom participants skip, Admin canonicalize):

cd modules/database
NODE_OPTIONS=--experimental-vm-modules ./node_modules/.bin/jest --testPathPatterns=src/__tests__/indexes --forceExit

Result: 5 suites, 92 passed. tsc --noEmit in modules/database is clean (after proto codegen).

Live-engine integration tests were not run in this environment (no Mongo/SQL). Recommended follow-up smoke:

• Compatible unique + descending compound on Mongo, PostgreSQL, MySQL/MariaDB, and SQLite, including re-register of a schema that already has engine-default names
• ChatMessage room / roomId compounds on PostgreSQL and MySQL
• ChatRoom participants still skipped
• Authz String *Id unchanged (resource stays resource)

Test Configuration:

• Unit tests via Jest + ts-jest (ESM)
• Node 24
• Workspace packages @conduitplatform/database, @conduitplatform/grpc-sdk

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published in downstream modules

Remaining risks

• showIndex row shapes differ by SQL dialect; unusual engine-specific indexes may normalize oddly.
• Compound Compatible DESC uses Sequelize { name, order } field objects — worth a live MySQL/MariaDB/SQLite check.
• Identity is fields + unique, not sort direction. Opposite-direction indexes on the same fields are treated as the same index.
• Admin create persist merges by name; stale generated metadata for an identity already live is cleaned on the next module re-register.
• Live SQL stores roomId while Admin/export now speak declared room. Older exports that used roomId are canonicalized on import when room is a scalar Relation.
• A declared Relation already named *Id would become *IdId (Sequelize relation + 'Id'). None of the platform models do that.