Optimize Settings::from_toml() call

[prk-Jr]

Summary

Optimize Settings::from_toml() by removing redundant config crate env var scanning at runtime. WASM guest profiling confirmed: ~4.8% → ~2.7% CPU per request.

Why this is safe

build.rs already resolves all configuration at compile time:

1. Reads trusted-server.toml
2. Merges TRUSTED_SERVER__* env vars via the config crate
3. Writes the fully-resolved result to target/trusted-server-out.toml
4. Registers cargo:rerun-if-env-changed for all possible env vars

At runtime, get_settings() loads this pre-resolved TOML via include_bytes!. The old from_toml() was re-running the entire config crate pipeline (env source scanning, config merging, deserialization) on every request — completely redundant since the embedded TOML already contains the final values. On Fastly Compute (WASM sandbox), there are no environment variables available at runtime anyway.

What changed

• from_toml() now uses direct toml::from_str() deserialization — no config crate, no env var scanning
• build.rs::merge_toml() now owns the env var merging logic inline using the config crate directly, instead of delegating to from_toml()
• config crate import in settings.rs is now #[cfg(test)] — not compiled into the production binary

Why some tests use from_toml_with_env()

build.rs runs once before tests start, producing a single static target/trusted-server-out.toml. But 7 tests need to verify that different TRUSTED_SERVER__* env vars correctly override different TOML values per test (e.g., overriding publisher.origin_url in one test, integrations.prebid.bidders in another). They can't rely on build.rs for this because:

• build.rs runs once, not per-test
• Each test sets different env vars via temp_env that build.rs never sees
• The resolved output file can't reflect per-test env var combinations

So these 7 tests call from_toml_with_env() directly — a #[cfg(test)] helper that preserves the original env var merging behavior. This lets them test the merging contract in isolation without going through build.rs.

All other tests (20+) use the new fast from_toml() and continue passing — they only test TOML parsing, not env var overrides.

Files changed

File	Change
settings.rs	from_toml() → toml::from_str; added #[cfg(test)] from_toml_with_env(); config import → #[cfg(test)]
build.rs	Env var merging logic moved here (was in from_toml())
Cargo.toml	Added toml to [dependencies]

Test plan

• cargo test — 352 tests pass
• WASM build succeeds
• WASM guest profiling confirms ~4.8% → ~2.7% CPU for get_settings()
• Deploy to staging, run BENCH_URL=<staging> ./scripts/benchmark.sh and compare against baseline
• Team review: verify build.rs env var merging matches original from_toml() behavior