ci: drop non-existent suites from the Linux and macOS test matrices

[ErikBPF]

Which issue does this PR close?

None. Small CI-hygiene follow-up; no tracking issue.

Rationale for this change

dev/ci/check-suites.py only verified that every suite in the source tree is
listed in a workflow. It never verified the reverse, so six suite names that no
longer exist remained in the linux-test and macOS matrices:

• org.apache.comet.CometExpressionCoverageSuite
• org.apache.comet.exec.CometAsyncShuffleSuite
• org.apache.comet.exec.DisableAQECometAsyncShuffleSuite
• org.apache.comet.parquet.ParquetReadV2Suite (removed in chore: Remove deprecated SCAN_NATIVE_COMET constant and related test code #3671)
• org.apache.spark.sql.comet.ParquetDatetimeRebaseV1Suite
• org.apache.spark.sql.comet.ParquetDatetimeRebaseV2Suite

CI is green only because the suites are passed via -DwildcardSuites, which
ignores unmatched names; an explicit -Dsuites list fails with
ClassNotFoundException. The matrices therefore claim coverage that does not
run, and nothing prevented a stale name from being re-added.

What changes are included in this PR?

• Add a forward declaration check to dev/ci/check-suites.py: a suite listed in
  a workflow must be declared as a class/trait/object in some sourceset.
• Remove the six dead suite lines from .github/workflows/pr_build_linux.yml
  and .github/workflows/pr_build_macos.yml.

ParquetReadV1Suite and CometStringDecodeSuite stay listed: the former is
declared in ParquetReadSuite.scala, the latter in the spark-3.x sourceset
used by the 3.4/3.5 profiles.

How are these changes tested?

• python3 dev/ci/check-suites.py reports the six undeclared names (RED) before
  the removal, then prints All workflow-listed suites are declared and exits 0
  after it.

[sunchao]

Summary

• Prior state and problem: Both CI matrices listed six nonexistent suites that wildcardSuites silently ignored.
• Design approach: Remove those entries and extend check-suites.py to check workflow names against Scala sources.
• Correctness / compatibility analysis: All six removed prefixes match no declarations in the checkout. Remaining matrix names have matching package declarations. Checked ScalaTest 3.2.16 discovery behavior and relevant Spark sources across supported 3.4–4.2 profiles. No introduced P1/P2 issues found within this review.
• Key design decisions: Scanning across source sets preserves version-specific suites and suites declared in differently named files. The implementation uses standard-library parsing without adding dependencies or unnecessary abstractions.
• Implementation sketch: Add declaration collection and undeclared-name diagnostics, then remove six entries from each workflow.
• Behavioral changes worth calling out: The checker now rejects the six stale entries. No active suite selection or Spark runtime behavior changes. The complete checker took approximately 0.36 seconds locally.
• Suggested improvements: None at the requested P1/P2 bar.

Reviewed the full three-file diff from 9a4d5f28368d4fbb54ac23c14c5d52d5637ce26d to f4810f0d836aa26f4723f5af7d5ce40b69d823a4. The PR is non-draft. Routed skill: review-comet-pr; no sibling area skills apply to this CI-only change. Snapshot and live discussion checks contained no reviews, comments, or threads.

Exact-head CI: label passed. Comet CI and CodeQL report action_required, with no jobs executed. Hosted build/test validation is therefore unavailable.

Local validation: check-suites.py, check-ci-config.py, and git diff --check passed. Disposable fixtures restored the base workflows and produced exactly 12 expected diagnostics, verified rejection of an unknown suite, and passed after restoring the head workflows. No full Maven/native build or Spark test suite was run. The checkout remains unchanged.

[ErikBPF]

Updated against pinned main 63fd1c9e. Rebase conflicts were resolved while retaining upstream matrix additions and CometUnboundedShuffleMemoryAllocatorSuite. Published ancestry is retained with an identical-tree merge, without force-pushing.

The reverse suite-declaration check remains: explicitly named matrix suites must resolve to real Scala suite declarations. Stale entries were removed from both Linux and macOS matrices; ParquetEncryptionITCase remains covered. There are 151 matrix entries per workflow, not 151 distinct Scala suites.

Verification on Apollo:

• The original stale matrices produced 12 expected diagnostics; corrected matrices passed.
• Deliberately invalid suite names produced the expected two diagnostics.
• CI configuration, syntax, diff checks and workflow-only actionlint passed; independent source review found no introduced defect.

Full actionlint still reports unchanged ShellCheck findings in unedited shell fragments (SC2006, SC2086, SC2129); it is not being reported as fully green. No Scala/runtime change was made in this PR.

Updated head: 81a7b21b53d1be6a81244a8b8bb278d91f6e3c2a (tested tree unchanged from cb525947eeef4e3c13bacca9a0446a1607ef0bf4).

Current-main diff correction

The intended patch was rebased onto ca1c143a. An ordinary merge of actual main preserved public ancestry without force-pushing and produced exactly the clean rebased tree. GitHub Files Changed is now verified at 3 files, +35/-12, limited to the two workflow matrices and check-suites.py.

Suite/CI guards, bad-name and guard-knockout controls, syntax and workflow-only actionlint passed on this base. Upstream float-suite additions, allocator naming and encryption IT remain intact. There are now 152 matrix entries per workflow; the earlier 151 count belonged to the older base. Full ShellCheck still has unchanged upstream diagnostics, not a fully green verdict.

Updated head: ad24e03bb9d314b69c6d478ce204f98a71df3225.

[andygrove]

Thanks @ErikBPF