build(deps): sea-query 1.0 + sea-query-sqlx 0.8

[shankar-gpio]

Supersedes #250 — that PR can be closed once this lands.

Why #250 was stuck

Bumping sea-query alone cannot resolve: sea-query-binder 0.7 pins sea-query ^0.32. Upstream superseded sea-query-binder with a renamed crate, sea-query-sqlx, which is what unblocks the upgrade.

Version choice

	from	to
sea-query	0.32.7	1.0.2
sea-query-binder → sea-query-sqlx	0.7.0	0.8.1

sea-query-sqlx 0.8.1, not 0.9.x, on purpose: 0.8.x is the line that still targets sqlx ^0.8 and declares rust-version = 1.88.0. So sqlx stays on 0.8 and the MSRV 1.88 promise in CLAUDE.md holds. 0.9 moves to sqlx 0.9 + Rust 1.94 — a separate, larger decision. Cargo says so itself while resolving: available: v0.9.1, requires Rust 1.94.0.

The port

Compiler- and clippy-driven throughout, not hand-hunted:

• Crate rename (31 sites). The binder type only ever surfaces behind storage::build_sqlx and the DbPool/DbTransaction execute helpers, so it is a rename and nothing else. build_any_sqlx was removed in 0.8 (1.0 replaced dyn QueryBuilder with impl QueryBuilder), but Orion never used it — build_sqlx already matches on DbBackend and calls the monomorphised builder.
• ExprTrait: comparison operators moved off Expr onto a trait, so 11 modules and 2 test files gained use sea_query::ExprTrait;. Without it .eq() resolves to PartialEq::eq, which is why the first check reported expected Expr, found bool.
• Ord::max vs ExprTrait::max: ExprTrait is blanket-implemented for everything convertible into an Expr, so a bare .max(0) on an i64 becomes ambiguous (E0034). Four sites now spell it Ord::max(a, b).
• Expr::cust takes Into<Cow<'static, str>>. trace_dlq threaded non-'static &str through it: now is only ever helpers::sql_now, so it is typed &'static str; lease_until arrives from sql_now_plus_secs as an owned String, so those two sites clone.
• SimpleExpr is now an alias of Expr and From<Expr> for SimpleExpr is gone, so 62 Expr::val(x).into() calls became identity conversions and tripped clippy::useless_conversion under -D warnings. Applied with cargo clippy --fix; it correctly left the real &str/bool → Expr conversions alone.
• Value variants are unboxed: one test assertion drops a Box::new.

schema.rs needed no change — all 11 Iden enums use the derive, which still snake_cases variants, and Iden::to_string survives, so the pinned column-identifier test holds. Nothing matches on ColumnRef/TableRef, so 1.0's AST rework and new #[non_exhaustive] are inert.

Deliberately not done: renaming SimpleExpr → Expr (the alias is not deprecated), and turning Alias::new(x) into bare x — Iden is implemented for &'static str only, and that path skips identifier escaping.

No golden SQL assertion changed. 1.0 renders byte-identical SQL for everything Orion builds, on all three dialects.

Dependencies

The graph shrinks: sea-query 1.0.2 has no mandatory dependencies, dropping inherent and postgres-types. sea-query-sqlx lists jiff-sqlx (Unlicense OR MIT) as optional behind with-jiff, which is off — so it lands in Cargo.lock but is never compiled. Noted in the manifest, since Unlicense is not in deny.toml's allowlist and it clears only via the MIT arm.

cargo deny: advisories, bans, licenses, sources — all ok.

Verification

Gate	Result
cargo fmt --all --check	ok
cargo clippy --workspace --all-targets -- -D warnings	clean
cargo test --workspace --all-targets	1091 lib + 657 integration, all pass
cargo test --doc	ok
cargo deny check	advisories/bans/licenses/sources ok
cargo package --locked --workspace	verify-builds all 4 crates
RUSTDOCFLAGS="-D warnings" cargo doc	clean
cargo +1.88 test --workspace	all pass — MSRV holds
storage_postgres / storage_mysql / schema_parity	6 / 4 / 1 pass
cluster (Docker)	17 pass
container-backed integration set	26 pass
tests/e2e/run.sh	75 pass

Postgres/MySQL rendering is only executed under Docker, so those suites were run rather than left to CI.

Two things for the reviewer

1. Kafka container tests are flaky at default parallelism on the machine this was verified on: failures are producer-side MessageTimedOut before any SQL runs, the count varied run to run (7, then 2), and all 20 pass with --test-threads=2. That is the contention ci.yml already predicts for this step, not a regression from this change — but the CI job runs without that flag, so it may need the flag added.
2. MSRV headroom is now zero: sea-query, sea-query-derive and sea-query-sqlx all declare exactly 1.88.0. Any future 1.0.x patch that raises rust-version will break the msrv job on the next cargo update.