Declare table-level indexes on entities with #[sea_orm(index(..))]

[LeenHawk]

Closes #3157.

Column-level attributes cannot express non-unique composite indexes, explicit column order, descending columns, or partial indexes. This adds table-level declarations so those indexes can live on the entity and participate in schema sync.

#[sea_orm::model]
#[derive(Clone, Debug, PartialEq, Eq, DeriveEntityModel)]
#[sea_orm(table_name = "post")]
#[sea_orm(
    index("idx_post_feed", on(author_id, created_at(desc))),
    index("uq_post_author_slug", on(author_id, slug), unique),
    index("idx_post_archived", on(archived_at), where = "archived_at IS NOT NULL")
)]
pub struct Model { /* ... */ }

Every index needs an explicit name. on(..) uses Rust field names and respects column_name / rename_all; unknown fields are compile errors. Columns can have (asc) or (desc).

where accepts either a SQL string or a SeaQuery expression implementing IntoCondition, for example where = Expr::col(Column::ArchivedAt).is_not_null() with sea_orm::ExprTrait in scope. Partial indexes are supported by SQLite and Postgres, not MySQL. NULL ordering is left out for now: SeaQuery's index API has no corresponding option.

The derive generates EntityName::table_indexes(). Its default is empty, so existing entities are unaffected. Schema::create_index_from_entity, the schema builder, and entity registry include these indexes.

Schema sync matches table-level indexes by name. It creates missing ones and keeps declared unique indexes; it does not compare definitions, so changing a definition requires a new name. Column-level matching is unchanged.

The macro's consume_meta also skips nested attribute lists, allowing #[sea_orm::model] and the other derives to pass over index(..).

Verified locally on Linux:

• cargo test --workspace
• DATABASE_URL="sqlite::memory:" cargo test --test schema_sync_tests --features default,sqlx-sqlite,runtime-tokio-rustls,schema-sync
• cargo clippy -p sea-orm -p sea-orm-macros --features sqlx-sqlite,runtime-tokio,schema-sync,macros -- -D warnings
• cargo +nightly fmt --all --check

Tests cover SQL generation for SQLite/Postgres/MySQL, both condition forms, renamed columns, invalid attributes, and repeated SQLite sync on an existing table. I haven't run the database integration tests against Postgres or MySQL locally.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[Huliiiiii]

1. I prefer syntax like this:

   #[sea_orm(
       index(
           "idx_post_feed",
           on(author_id, created_at(desc, nulls_last))
       ),
       index(
           "uq_post_author_slug",
           on(author_id, slug),
           unique
       ),
       index(
           "idx_post_archived",
           on(archived_at),
           where = "archived_at IS NOT NULL" // or
           where = Expr::col(Column::ArchivedAt).is_null() // sea query expr / impl IntoCondition
       )
   )]

2. We should require an explicit name for index, as it would improve maintainability and compatibility.

[LeenHawk]

Thanks, the on(..) syntax reads better. Updated in 1297c31: names are now required, columns use on(author_id, created_at(desc)), and where accepts either a SQL string or a SeaQuery expression/condition. I removed the automatic naming path; sync still matches these indexes by name.

One part I left out is nulls_last / nulls_first. SeaQuery 1.0.2's index API only has Asc and Desc, with no NULL ordering option. That would need SeaQuery support first rather than a SQL workaround here.

Workspace tests, the SQLite sync tests, Clippy, and formatting pass locally. Updated the examples and PR description too.