ci: keep the codspeed-macro turbo cache apart from GitHub's arm runners

The codspeed-macro jobs had no cache key suffix, so they restored Turbo
caches written by the arm64 `check` job on GitHub's runners. Turbo does not
hash the host's libstdc++, so the replayed core prebuild required a newer
CXXABI than the macro runner provides and failed to load.

This only surfaced once an example installed core as a copy: in the
workspace, core loads the addon compiled by the root install instead of the
prebuild. Give both codspeed-macro jobs their own suffix so they share a
cache with each other only.

Refs COD-2931
Co-Authored-By: Claude <noreply@anthropic.com>

Author: GuillaumeLagrange

.github/workflows/codspeed.yml

@@ -82,6 +82,10 @@ jobs:
           node-version-file: .nvmrc
       - name: Restore turbo cache
         uses: ./.github/actions/turbo-cache
+        # The native addon prebuild links against the host's libstdc++, which
+        # Turbo does not hash; keep GitHub's arm64 runners' builds out.
+        with:
+          key-suffix: codspeed-macro
       - run: pnpm install --frozen-lockfile --prefer-offline
       - run: pnpm turbo run build
       - name: Install vitest examples
@@ -154,6 +158,8 @@ jobs:
 
       - name: Restore turbo cache
         uses: ./.github/actions/turbo-cache
+        with:
+          key-suffix: codspeed-macro
 
       - name: Install repo dependencies
         run: pnpm install --frozen-lockfile --prefer-offline