Keep sccache server alive during long jobs

.gitea/actions/rust-cache/action.yml

@@ -97,6 +97,7 @@ runs:
         echo "enabled=true" >> "$GITHUB_OUTPUT"
         echo "RUSTC_WRAPPER=sccache" >> "$GITHUB_ENV"
         echo "SCCACHE_IGNORE_SERVER_IO_ERROR=1" >> "$GITHUB_ENV"
+        echo "SCCACHE_IDLE_TIMEOUT=3600" >> "$GITHUB_ENV"
         echo "SCCACHE_LOG=${SCCACHE_LOG:-warn}" >> "$GITHUB_ENV"
         if [ -n '${{ inputs.sccache-bucket }}' ]; then
           repo="${GITHUB_REPOSITORY:-unknown/repository}"

README.md

@@ -68,4 +68,7 @@ Omit `sccache-bucket` to use the runner-local sccache cache:
   variables from workflow secrets.
 - `SCCACHE_IGNORE_SERVER_IO_ERROR=1` is set so a cache outage falls back to
   local compilation instead of failing the build.
+- `SCCACHE_IDLE_TIMEOUT=3600` keeps the server alive across long build/test
+  steps, which matters when the setup step has S3 credentials scoped only to
+  that step.
 - Add `sccache --show-stats` after build/test steps to inspect cache hit rates.