ci: collect LLGo test coverage with Codecov

[cpunion]

The LLGo workflow now collects and uploads coverage from the full test suite on macOS and Ubuntu. The workflow uses xgo-dev/setup-llgo with llgo-version: main, Go 1.27 and LLVM 22. The workflow uses the published setup-llgo v0.2.0 release, which installs the required linker and libclang dependencies. Its implementation and review fixes were merged in xgo-dev/setup-llgo#50. The repository-local setup action is unchanged.

The test command uses -coverpkg=./... so calls from the cl and tool tests also measure local dependencies such as clang, cl/cltest, and tool/listth. An explicit empty Codecov ignore list includes every reported package, including tool. Only the LLGo-generated coverage.txt is uploaded, with automatic report discovery disabled; go tool cover only reads that profile.

Validation:

• llgo test -v -coverpkg=./... -covermode=atomic -coverprofile=coverage.txt ./... passed locally with LLGo main at 7db1409073f28bf13a35ae0a3ea663eba010fd73; go tool cover reported 80.1% statement coverage.
• CI run 36294357536 using the published xgo-dev/setup-llgo@v0.2.0 release passed the full LLGo tests and Codecov uploads on macOS and Ubuntu. Statement coverage was 80.1% and 70.2%, respectively.
• Codecov merged the two uploads into a report with 28 files across all six current top-level source directories and 79.11% line coverage.
• actionlint .github/workflows/llgo.yml, git diff --check, and Codecov YAML validation passed.

Codecov's project check remains below the 83.32% baseline from 55f289c, which predates the repository rewrite and is 393 commits behind the PR base. The old report used go test without -coverpkg and covered a different source tree. Coverage thresholds are unchanged.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.11%. Comparing base (55f289c) to head (b69078e).
⚠️ Report is 393 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #854      +/-   ##
==========================================
- Coverage   83.32%   79.11%   -4.21%     
==========================================
  Files          28       28              
  Lines        2500     1743     -757     
==========================================
- Hits         2083     1379     -704     
+ Misses        369      364       -5     
+ Partials       48        0      -48     

Flag	Coverage Δ
llgo-tests	79.11% <ø> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[fennoai]

Review: switch LLGo to build-from-main + Codecov coverage

The change is well-structured overall. Path wiring is consistent: go -C .llgo build -o bin/llgo resolves to .llgo/bin/llgo, matching the release-tarball layout, and PATH ($GITHUB_WORKSPACE/.llgo/bin) + LLGO_ROOT ($GITHUB_WORKSPACE/.llgo) line up for both code paths. The Codecov step is well-guarded (github.repository check, !cancelled(), hashFiles presence), and use_oidc correctly restricts OIDC to same-repo events.

Note: xgo-dev/llgo is not new — download-llgo.sh already uses it as the release source, so it is the intended upstream, not a deviation introduced here.

Main themes below: (1) tracking an unpinned main on every run has supply-chain, reproducibility, and CI-cost consequences; (2) the from-source build is now on the hot path uncached and CPU-throttled; (3) a couple of small cleanups.

[fennoai]

Review (cont.): two findings from the switch to build-from-main

Follow-up to the inline comments above — these two were on the build step and are the highest-value points.