C++: Fix join order in definitionHasPhiNode

[MathiasVP]

When trying to make sense of https://github.com/github/codeql-c-analysis-team/issues/190 yesterday, I noticed a bad join order (which sadly didn't fix that issue, but nonetheless seems worth fixing) on kamailio/kamailio.

Before:

[2020-11-26 09:49:25] (682s) Starting to evaluate predicate SSAConstruction::PhiInsertion::definitionHasPhiNode#2#ff/2@i6#72c4bw (iteration 6)
[2020-11-26 09:49:27] (683s) Tuple counts for SSAConstruction::PhiInsertion::definitionHasPhiNode#2#ff/2@i6#72c4bw:
    23942751 ~1%      {3} r1 = JOIN SSAConstruction::PhiInsertion::definitionHasDefinitionInBlock#2#ff#prev_delta AS L WITH SSAConstruction::PhiInsertion::definitionLiveOnEntryToBlock#2#ff AS R ON FIRST 1 OUTPUT L.<1>, R.<1>, L.<0>
    11753    ~24%     {2} r2 = JOIN r1 WITH Dominance::getDominanceFrontier#2#ff AS R ON FIRST 2 OUTPUT r1.<2>, r1.<1>
    3224     ~36%     {2} r3 = r2 AND NOT SSAConstruction::PhiInsertion::definitionHasPhiNode#2#ff#prev AS R(r2.<0>, r2.<1>)
    return r3

After:

[2020-11-26 09:22:05] (652s) Starting to evaluate predicate SSAConstruction::PhiInsertion::definitionHasPhiNode#2#ff/2@i6#64968w (iteration 6)
[2020-11-26 09:22:05] (652s) Tuple counts for SSAConstruction::PhiInsertion::dominanceFrontierOfDefinition#2#ff/2@i5#64968x:
    320824  ~0%      {2} r1 = SCAN SSAConstruction::PhiInsertion::definitionHasDefinitionInBlock#2#ff#prev_delta AS I OUTPUT I.<1>, I.<0>
    314862  ~17%     {2} r2 = JOIN r1 WITH Dominance::getDominanceFrontier#2#ff AS R ON FIRST 1 OUTPUT r1.<1>, R.<1>
    242301  ~7%      {2} r3 = r2 AND NOT SSAConstruction::PhiInsertion::dominanceFrontierOfDefinition#2#ff#prev AS R(r2.<0>, r2.<1>)
    return r3
...
[2020-11-26 09:22:05] (652s) Tuple counts for SSAConstruction::PhiInsertion::definitionHasPhiNode#2#ff/2@i6#64968w:
    13051    ~0%     {2} r1 = JOIN SSAConstruction::PhiInsertion::dominanceFrontierOfDefinition#2#ff#prev_delta AS L WITH SSAConstruction::PhiInsertion::definitionLiveOnEntryToBlock#2#ff AS R ON FIRST 2 OUTPUT L.<0>, L.<1>
    return r1

I can't quite work out why the intermediary tuple counts have changed, but the size of the final predicate is unchanged:

Before:

[2020-11-26 09:49:27] (684s)  >>> Created relation SSAConstruction::PhiInsertion::definitionHasPhiNode#2#ff/2@72c4bw with 404674 rows.

After:

[2020-11-26 09:22:05] (652s)  >>> Created relation SSAConstruction::PhiInsertion::definitionHasPhiNode#2#ff/2@64968w with 404674 rows.

CPP-differences: https://jenkins.internal.semmle.com/job/Changes/job/CPP-Differences/1572/

[geoffw0]

Code change LGTM. Waiting to confirm performance on the CPP-Differences job.

[MathiasVP]

Performance numbers look relatively unchanged. The numbers might be a slight improvement, but it's all within the noise range.