Run build first.

Run individual tests like

    pg15_tests/test_D29226.sh

Run all tests once like

    pg15_tests/run_tests.sh

run_tests.sh can alternatively take test specs as input, where a test spec is
of the form <program><tab><descriptor>[<tab><flaky>]:

    # Run passing tests once:
    pg15_tests/run_tests.sh <pg15_tests/passing_tests.tsv
    # Run flaky tests once:
    pg15_tests/get_flaky_test_specs.sh | pg15_tests/run_tests.sh
    # Run shell tests once:
    pg15_tests/get_shell_test_specs.sh | pg15_tests/run_tests.sh
    # Run the first flaky test:
    pg15_tests/get_flaky_test_specs.sh | head -1 | pg15_tests/run_tests.sh
    # Run the first flaky test n times:
    for _ in {1..10}; do pg15_tests/get_flaky_test_specs.sh | head -1; done | pg15_tests/run_tests.sh
    # Run any tests that historically failed:
    grep -v '^0' build/latest/pg15_test/results.tsv | cut -f 3,4,5 | pg15_tests/run_tests.sh
    # Same thing but deduplicated:
    grep -v '^0' build/latest/pg15_test/results.tsv | cut -f 3,4,5 | sort -u | pg15_tests/run_tests.sh

Add passing tests into passing_tests.tsv.  Add <5% flaky tests into
flaky_tests.tsv.  >5% flaky and failing tests can be tracked in shell tests.
Add shell tests by adding a new test_*.sh file following the existing
conventions.

Make sure new tests pass 10x in a row to avoid introducing flaky tests, and
make sure all tests pass (at least once) to avoid regressions.  Note: it may be
more convenient to use jenkins to cover passing and flaky tests:

    pg15_tests/get_shell_test_specs.sh | pg15_tests/run_tests.sh
    # ...and trigger jenkins

In the later stages of pg15, there will be no more need for these tests because
the underlying tests will pass.

(When, in the future, deactivating these tests, don't forget to remove the line
in .arclint.)