CI: run every job on spark-1; drop rust-cache

Two failures, two causes.

GitHub-hosted jobs never started: 'recent account payments have failed or
your spending limit needs to be increased'. Not a config problem. Public
repos get hosted runners free, so this may resolve itself; meanwhile every
job runs on spark-1, which is also the deployment architecture (Oracle
Ampere A1 is aarch64), so this is an improvement rather than a workaround.

The aarch64 job failed with exit 127 — cargo not found — and kept failing
after each fix. Cause: Swatinem/rust-cache's post-step prunes ~/.cargo.
That is correct on a GitHub-hosted runner where ~/.cargo is disposable, and
destructive on a self-hosted one where it is the real toolchain. It deleted
~/.cargo/bin/rustup after the first successful run, leaving every shim a
dangling symlink. The action is removed and the toolchain moved to
/opt/rust, wired through the runner's .env and .path, so it is out of reach
even if someone re-adds it. Caching bought nothing here anyway — the runner
keeps its target dir between runs.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JkyvfNJGTshJNE9FtwPLk7

.github/workflows/ci.yml

@@ -14,40 +14,49 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
 
+# Every job runs on spark-1 (aarch64, 20 cores). Two reasons, and the second
+# outlives the first:
+#
+#  1. GitHub-hosted runners are currently unavailable on this account for
+#     billing reasons. Public repos get them free, so this may be temporary.
+#  2. aarch64 is the deployment target — Oracle Cloud's Ampere A1. Testing on
+#     the architecture we ship is worth more than testing on x86 and hoping.
+#
+# An x86_64 job is worth adding back once hosted runners work, to catch
+# endianness- and width-dependent bugs. Tracked in ROADMAP.md.
+#
+# ⚠️ Do NOT add Swatinem/rust-cache to these jobs. On a self-hosted runner its
+# post-step prunes ~/.cargo — which here is a real, persistent toolchain, not a
+# disposable one. It silently deleted the rustup binary and every subsequent
+# job failed with exit 127. The toolchain now lives at /opt/rust (wired via the
+# runner's .env and .path) partly to put it out of reach, but do not re-add the
+# action. The runner keeps its target dir between runs anyway, so caching buys
+# nothing here.
+
 jobs:
   check:
     name: check · test · clippy · fmt
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, ARM64, spark-1]
     steps:
       - uses: actions/checkout@v5
-      - uses: dtolnay/rust-toolchain@stable
-        with:
-          components: rustfmt, clippy
-      - uses: Swatinem/rust-cache@v2
       - run: cargo fmt --all --check
       - run: cargo clippy --workspace --all-targets --all-features
       - run: cargo test --workspace --all-features
 
   licences:
     name: licence policy (ADR 0002)
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, ARM64, spark-1]
     steps:
       - uses: actions/checkout@v5
-      - uses: EmbarkStudios/cargo-deny-action@v2
+      - run: cargo deny check licenses bans sources advisories
-        with:
-          command: check licenses bans sources advisories
 
-  # Native aarch64 build on spark-1. Oracle Cloud's free tier is Ampere A1
+  release:
-  # (aarch64), so the release artefact must be aarch64 — and spark-1 (GB10,
+    name: release build · aarch64
-  # 20 cores, 121 GB) builds it natively rather than cross-compiling.
-  aarch64:
-    name: release build · aarch64 (spark-1)
     runs-on: [self-hosted, Linux, ARM64, spark-1]
     if: github.ref == 'refs/heads/main' || github.event_name == 'workflow_dispatch'
     steps:
       - uses: actions/checkout@v5
-      - uses: Swatinem/rust-cache@v2
+      - run: cargo build --release --workspace --locked
-      - run: cargo build --release --workspace
       - uses: actions/upload-artifact@v4
         with:
           name: openmail-aarch64