GeezerVison

Some environments cannot run cargo when a pipeline starts. A
deterministic simulation test runs on a single slow CPU with no crate
registry, yet still needs to start pipelines for a known set of SQL
programs.

The compiler server gains --prebuilt-binaries-directory. Before running
cargo, it looks there for a binary of the same program and, if one
exists, delivers it instead. A binary is valid for any pipeline with the
same SQL, but the source checksum cannot find it: it hashes the encoded
generated code, which carries file modification times, and the
generated crate names embed the pipeline id. The lookup key is
therefore the source checksum with the generated code hashed by file
paths and contents, under the crate names of the nil pipeline.

--precompile fills the directory: each --prebuild-sql-program file is
compiled as the nil pipeline, like the dummy program, and its binary is
copied there under its lookup key.

The compiler now also passes CARGO_BUILD_TARGET to cargo and reads
artifacts from target/<triple>/. With an explicit target, RUSTFLAGS
apply only to target artifacts, so coverage instrumentation flags no
longer break the linking of build scripts and proc macros.

Signed-off-by: Gerd Zellweger <[email protected]>

… key

find_prebuilt_binary receives the prebuilt binary key, which differs
from the source checksum its parameter and test were named after.

Signed-off-by: Gerd Zellweger <[email protected]>

Update the test-only PostgreSQL connector to the current TLS API and
use the Rust 2024-compatible random helper. This lets prebuilt cache
unit tests compile without installing an embedded database.

Blind namespace replacement can change SQL literals and conflate distinct
programs. Normalize generated Rust identifiers and Cargo crate references
structurally, including workspace members, while preserving literal data.
Reject duplicate normalized archive paths instead of overwriting them.

Cover namespace independence, literal collisions, Cargo references and
malformed generated sources. The four prebuilt cache tests pass, and real
SQL compiler output normalizes identically across pipeline IDs.