fix(ci): remove the three sources of flakiness in the weekly example runs

[marc0olo]

Three unrelated causes behind the weekly-run failures, each reproduced before fixing.

evm_block_explorer (not flaky — reliably broken)

test.sh asserted on Ethereum mainnet block 1. PublicNode routes requests across regional node pools and some of them only retain history from shortly before the Merge, so the call comes back as pruned history unavailable: requested 1, earliest available 15500000 depending on which pool serves the outcall. Both the Motoko and Rust jobs failed on it, and re-running does not help from a region whose pool is pruned.

Now asserts on block 20000000 (post-Merge, ~4.5M blocks above that boundary). Verified locally: all three assertions pass. The READMEs note the limitation so users who query early blocks in the frontend know why it fails.

unit_testable_rust_canister (flaky)

test_get_proposal_info and test_get_proposal_titles both call get_governance_wasm(), which downloaded to a single shared path with no synchronisation. libtest runs them on separate threads, so std::fs::write truncating the file under the peer's std::fs::read yields a partial gzip — surfacing as the opaque CanisterInvalidWasm: unsupported canister module format in the failed run. curl's exit status was also never checked, so a failed transfer wrote an empty file that the exists() check would then treat as a valid cached artifact.

Each artifact is now prepared exactly once per test binary behind a LazyLock, the download is status-checked with retries, and it lands via a temp file + rename so a partial transfer is never cached.

parallel_calls / pub-sub — Motoko jobs only (flaky, upstream bug)

Both are the only Motoko examples with two canisters in one mops project. icp-cli runs one mops build <canister> per canister concurrently, and on a cold cache both invocations download and extract the moc toolchain into the same two shared locations — .mops/_tmp/motoko-Linux-x86_64-<ver>.tar.gz and ~/.cache/mops/moc/<ver>. Whichever build loses reads a half-written archive or execs a half-extracted moc, and the failure surfaces as Build failed for canister <name> (exit code: undefined) with no diagnostics — or, when the timing lands differently, as the ZlibError: zlib: unexpected end of file visible in run 32750148987.

That matches which canister fails being random across runs (caller, callee, publisher, subscriber have all appeared) and only ever in these two projects.

Reproduced locally with icp build on motoko/pub-sub, 3/3 failures on a cold moc cache, and still present on the latest mops (3.2.0). It is a mops bug, filed as caffeinelabs/mops#818 with the offending code path and a minimal reproduction. mops install up front makes the download happen once, which is what this adds to _run-example.yml; 4/4 clean afterwards. Guarded on mops.toml existing and mops being on PATH, so it is a no-op for Rust examples, and to be removed once the upstream fix ships.

🤖 Generated with Claude Code

[Copilot]

🟡 Changes recommended

Truncated gzip cache entries must be fully validated before reuse.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

This PR reduces weekly CI flakiness across EVM, Rust, and Motoko examples.

Changes:

• Uses retained Ethereum block 20,000,000 and documents pruning limitations.
• Synchronizes and hardens Rust WASM downloads.
• Pre-warms Motoko toolchain caches in CI.

File summaries

File	Reviewed changes
rust/unit_testable_rust_canister/backend/tests/integration_tests.rs	Synchronizes artifact preparation and adds reliable downloads. Moderate finding: truncated gzip cache entries are not fully validated.
rust/evm_block_explorer/test.sh	Tests a retained Ethereum block.
rust/evm_block_explorer/README.md	Documents PublicNode history limitations.
motoko/evm_block_explorer/test.sh	Tests a retained Ethereum block.
motoko/evm_block_explorer/README.md	Documents PublicNode history limitations.
.github/workflows/_run-example.yml	Pre-warms the Motoko toolchain cache.

Review details

• Files reviewed: 6/6 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟢 Approval recommended

The reviewed changes address the identified CI flakiness without unresolved blocking issues.

Review details

• Files reviewed: 7/7 changed files
• Comments generated: 0 new
• Review effort level: Lite