Run tests in parallel in weekly CI

[Naviabheeman]

Last weekly CI job ended in timeout as Github action can run only for 6 hours. Change the job a little to improve parallelism:

• Collapsed feature-pruning/feature-dbcrash into one heavy-tests job with strategy.matrix — build pipeline lives once, matrix-driven test_args/name differentiate the two.
• Added step to Print full combined logs on failure: runs combine_logs.py directly over every test tmpdir and writes the full, untruncated output (all nodes + test framework log) to — works even for a Timeout result, which test_runner.py itself never surfaces to the console.
• Retained the dead-timeout removal, chmod cleanup, ulimit fix, and dangling-process kill from the earlier split.

[azuchi]

Matrix split shape is right. A few items worth resolving before merge:

Recommended (in-scope)

1. if: failure() on the combined-logs step won't fire on the failure mode this PR targets. A job hitting the 6h runner ceiling terminates with conclusion: cancelled, not failure, so the log-dump step is skipped in exactly the case that motivated it. Use if: failure() || cancelled().

2. $GITHUB_STEP_SUMMARY has a 1 MB per-step cap. combine_logs.py on a heavy test easily produces tens of MB and GitHub silently drops the excess, so the "full untruncated" claim doesn't hold in practice. Either tail -c 950000 before appending, or write the full log to a file and upload it as an artifact with the summary linking to it.

3. concurrency.group: ${{ github.workflow }}-heavy (line 28) is a single global slot. Same shape as daily-test.yml's old group that PR Daily smoke benchmark ci #440 fixed to ${{ github.workflow }}-${{ github.ref }}. Weekly contention is rarer but a workflow_dispatch during a schedule run still cancels the earlier one unnecessarily. Apply the same one-line fix here for consistency.

Nit (fine as follow-up)

4. Debug core dump (on failure) is the pre-Daily smoke benchmark ci #440 shape — single most-recent dump only, hardcoded tapyrusd binary, no summary output. PR Daily smoke benchmark ci #440 rewrote the daily version to loop over every core.*, resolve the crashing binary from file output, and emit bt full into $GITHUB_STEP_SUMMARY. Copy that block verbatim.

5. HEAVY_SCRIPTS in test_runner.py:255 is now a second source of truth for "what runs weekly." Before: workflow ran --heavy, list lived in one place. After: the matrix drives it, and HEAVY_SCRIPTS is only kept alive by the completeness check at test_runner.py:663. Add a one-line comment on HEAVY_SCRIPTS saying "kept for the completeness check only; the weekly workflow drives what actually runs" so a future editor knows both places need updating together.

Also — the weekly workflow has no runs on this SHA (only schedule + workflow_dispatch triggers). A workflow_dispatch against this branch before merging would exercise the matrix split + combine_logs.py path for real.