SRE-3704 ci: adapt NLT pipeline to cover Fault Injection testing procedure.

[grom72]

This PR introduces logic that simplifies Fault Injection testing stage setup in the Jenkinsfile.
Requires:

• https://github.com/daos-stack/system-pipeline-lib/pull/23

NLT Fault Injection testing is fully implemented using the unitTest/unitTestPost Groovy
procedures from pipeline-lib:

    steps {
        job_step_update(
            unitTest(timeout_time: 600,
                     inst_repos: daosRepos(),
                     test_script: 'ci/unit/test_nlt.sh --memcheck no' +
                                  ' --system-ram-reserved 4 --server-debug WARN' +
                                  ' --log-usage-import nltr.json' +
                                  ' --log-usage-save nltr.xml' +
                                  ' --class-name fault-injection fi',
                     unstash_opt: true,
                     unstash_tests: false,
                     inst_rpms: unitPackages(target: 'el9') + ' daos-client-tests',
                     image_version: 'el9.7'))
    }
    post {
        always {
            unitTestPost artifacts: ['nlt_logs/'],
                         testResults: 'nlt-junit.xml',
                         always_script: 'ci/unit/test_nlt_post.sh',
                         valgrind_stash: 'fault-inject-valgrind',
                         nlt_name: 'Fault injection issues'
            ...
        }
    }

The stage is given a new name (NLT Fault Injection testing) so it is not confused with
the existing Fault injection testing stage.

parseStageInfo.groovy:

• Detect the NLT Fault Injection stage by name (case-insensitive) and set the FI flag.
• When FI is set, skip valgrind configuration — NLT FI runs with --memcheck no and
  does not produce memcheck XML files.
• When FI is not set (plain NLT), keep the existing valgrind_pattern/with_valgrind
  setup unchanged.

unitTest.groovy:

• Pass environment: "VM_CPUS=20" to provisionNodes for all NLT stages; the NLT test
  suite requires at least 20 CPU cores to run reliably.
• Move the results stash into a finally block so ignore_failure is always written to
  the stash even if afterTest() throws. Without this, unitTestPost() reads the earlier
  stash from runTest() which lacks the ignore_failure key, causing
  allowEmptyArchive: null and a NullPointerException in ArtifactArchiver.
• Remove config['NLT'] from the Valgrind check condition — NLT FI does not run with
  memcheck, so the Valgrind path should only be taken when with_valgrind is set.
• Guard Valgrind tarball creation with a fileExists() check to avoid a shell error when
  the copy step produced no output directory.

unitTestPost.groovy:

• Replace all direct results['ignore_failure'] map accesses with
  .get('ignore_failure', false) to prevent NPE when the key is absent in the stash.
• Replace the single tool: with a tools: list (nltTools) so the Fault Injection
  stage can report both nlt-errors.json and nlt-client-leaks.json as separate issue
  sources under recordIssues.
• Set recordIssues name: dynamically to 'Fault injection' or 'NLT' based on the
  FI flag, replacing the hard-coded 'Node local testing' label.
• Archive the memcheck tarball directly via archiveArtifacts rather than appending it
  to the deferred artifact list, consistent with how other binary artifacts are handled.
• Remove the NLT condition from the with_valgrind block — same rationale as unitTest.

skipStage.groovy:

• Add 'NLT Fault injection testing' as a recognized case so the new stage name is
  handled identically to 'Fault injection testing' when evaluating skip conditions.

[daltonbohning]

It appears this goes along with daos-stack/daos#17953.

So we are effectively going to have different handling for this stage based on the release branch? Since the stage name is different, pipeline-lib will treat it differently