Fix redshift recipe: correct the community-build accelerator error message

[claudespice]

What the recipe said

redshift/README.md Step 3 tells readers what a stock-installed runtime does with the write spicepod's engine: postgresql:

a stock-installed runtime rejects it with Unknown engine: postgres

What the code does

Unknown engine: {engine} exists, but it is not the message a reader hits here. It lives in DataAccelerator::create_accelerated_table (crates/runtime/src/dataaccelerator/mod.rs#L266-L271) — reached only after dataset init, which performs its own availability check first and fails there:

// crates/runtime/src/init/dataset.rs#L1312-L1320
self.accelerator_engine_registry
    .get_accelerator_engine(acceleration_settings.engine)
    .await
    .context(AcceleratorEngineNotAvailableSnafu {
        name: accelerator_engine.to_string(),
    })?;

engine: postgresql parses successfully — Engine::try_from accepts both postgres and postgresql (crates/runtime-acceleration/src/engine.rs#L57-L73) — so the engine is never "unknown". It is simply not registered when the binary is built without postgres-accel, and AcceleratorEngineNotAvailable is cfg-gated to render the community variant (crates/runtime/src/lib.rs#L318-L328):

#[cfg(not(feature = "postgres-accel"))]
#[snafu(display(
    "The accelerator engine {name} is not available. Valid engines are arrow, cayenne, duckdb, and sqlite."
))]
AcceleratorEngineNotAvailable { name: String },

Engine::PostgreSQL renders as postgres via Display, so the reader sees:

The accelerator engine postgres is not available. Valid engines are arrow, cayenne, duckdb, and sqlite.

This also matches what postgres/accelerator/README.md already says, so the two recipes now agree.

What changed

Quoted the actual error message. The rest of the note is unchanged and verified against v2.1.1: postgres-accel is still a real, non-default spiced feature (bin/spiced/Cargo.toml#L152), and release and models in the suggested cargo run line both exist.

Static verification only — confirming at runtime would require provisioning a Redshift cluster.