fix: harden domain reliability, concurrency, and runtime structure

.github/workflows/ci-backend.yml

@@ -20,13 +20,24 @@ jobs:
   backend:
     runs-on: ubuntu-latest
     timeout-minutes: 20
+    env:
+      RUST_LOG: "off"
+      # This workspace links ~24 separate integration-test binaries against a
+      # heavy dependency tree (tokio, hyper, rustls, opentelemetry,
+      # aws-lc-sys, ...) -- lld is meaningfully faster than the default bfd
+      # linker for that many large links, with no behavior change. Scoped to
+      # CI only; local dev toolchains are untouched.
+      RUSTFLAGS: "-C link-arg=-fuse-ld=lld"
     steps:
       - name: Checkout repository
         uses: actions/checkout@v7
 
       - name: Update Rust toolchain
         run: rustup update stable
 
+      - name: Install lld linker
+        run: sudo apt-get update && sudo apt-get install -y lld
+
       - name: Cache Rust dependencies
         uses: Swatinem/rust-cache@v2
         with:
@@ -47,18 +58,17 @@ jobs:
           cntryl-tools validate-benchmarks --config .cntryl/repository.toml
           cntryl-tools check-module-sizes --config .cntryl/repository.toml
 
-      - name: Cargo build
-        run: cargo build --locked
+      - name: Cargo check
+        run: cargo check --locked
 
       - name: Cargo clippy pedantic
         run: cargo clippy --locked --workspace --all-targets --all-features -- -D warnings -D clippy::pedantic
 
-      - name: Cargo test
+      - name: Build tests
+        run: cargo test --no-run --locked --workspace
+
+      - name: Run tests
         env:
-          # Shared CI runners have been observed running the e2e suites roughly
-          # two orders of magnitude slower than a developer machine, tripping
-          # the fixed per-frame deadlines the tests pass. Scale them here rather
-          # than loosening the deadlines for everyone.
           FITZ_TEST_TIMEOUT_MULTIPLIER: "10"
         run: cargo test --locked --workspace