fix(codegen): make dynamic-slice queries compile on sqlx 0.9

[jrandolf]

Implements the first part of the stack proposed in Discussion #47. Thanks to @iamralch for finding the bug and writing the fix. This carries their commit unchanged from iamralch#3.

Problem

The workspace targets sqlx 0.9, but a sqlc.slice() query that cannot use the = ANY($1) fast path emits sqlx::query_as(&sql) for its run-time SQL. sqlx 0.9 only accepts a &'static str directly, so that output does not compile. Reproduced against sqlx 0.9.0 with the code main currently generates for tests/e2e/cases/dynamic_slice:

error[E0277]: dynamic SQL strings should be audited for possible injections
  --> src/queries.rs:74:62

The e2e fixtures pinned sqlx 0.8, and no example had a non-ANY slice, so nothing compiled this output against the version the workspace declares.

Change

• The nine dynamic-SQL call sites in query.rs and batch.rs now pass sqlx::AssertSqlSafe(sql).
• The e2e fixtures move from sqlx 0.8 to 0.9.
• The README states the supported sqlx version.
• Only the two dynamic-slice snapshots change.

Why AssertSqlSafe is sound

The run-time string starts from the generated const SQL and changes in only two ways: placeholders are renumbered to $n (or expanded to ?), and an empty slice becomes NULL. Only the slice's length is decided at run time. Every value still goes through .bind(), so no caller data reaches the SQL text.

Support note

Generated code that contains a dynamic slice now needs sqlx 0.9. It no longer compiles against 0.8. Static queries are unchanged and still compile on both versions. The workspace already declared 0.9. If we treat dropping 0.8 for dynamic slices as breaking, retitle with ! before squashing so release-please records it.

Verification

• cargo fmt --check, cargo clippy --workspace --all-features --all-targets -D warnings, and cargo package --locked are clean.
• 88 lib tests and 30 codegen snapshot tests pass.
• The e2e suite passes against PostgreSQL with sqlx 0.9.
• An extra check (not committed) compiled every dynamic-slice command on sqlx 0.9 and ran it against PostgreSQL 17 with correct results: :one, :many (including an empty slice), :exec, :execrows, :execresult, :execlastid, :batchexec, :batchone and :batchmany.

The MySQL and SQLite parts of the stack (iamralch#1, iamralch#2) are not included. Review follows in the Discussion.