Unnest compound mode search loops and clean up inter search

[kslu-aom]

This change includes the following cleanups

• Unnest JMVD scale factor, CWP, and RefineMV loops in handle_compound_inter_prediction to avoid redundant motion search and skip invalid tool combinations. The new search order is 1) base mode, 2) refine MV, 3) CWP, 4) JMVD scale. This search order is intended to allow potential early termination in the future
• Remove unused variables, unnecessary checks, redundant mbmi field re-assignments, and move the update of common variable outside of search loops.
• Use reset_inter_mode_info helper to unify mode_info field resets

STATS_CHANGED

Results tested with RA 33 frames based on ae462ed

    1) Speed 0 (partial)
      +------------+-------+-------+-------+-------+-------+-------+
      | Class      |     Y |    Cb |    Cr |  wAvg |  Enc% |  Dec% |
      +------------+-------+-------+-------+-------+-------+-------+
      | A1         | -0.02 | -0.03 | +0.04 | -0.01 | 100.1 |  99.3 |
      | A2         | -0.00 | -0.03 | -0.04 | -0.01 |  99.9 |  99.8 |
      +------------+-------+-------+-------+-------+-------+-------+
    
    2) Speed 1
      +------------+-------+-------+-------+-------+-------+-------+
      | Class      |     Y |    Cb |    Cr |  wAvg |  Enc% |  Dec% |
      +------------+-------+-------+-------+-------+-------+-------+
      | A1         | +0.01 | +0.02 | +0.03 | +0.01 |  99.3 |  99.0 |
      | A2         | +0.01 | +0.01 | -0.06 | +0.00 |  99.8 |  99.3 |
      +------------+-------+-------+-------+-------+-------+-------+
    
    3) Speed 2
      +------------+-------+-------+-------+-------+-------+-------+
      | Class      |     Y |    Cb |    Cr |  wAvg |  Enc% |  Dec% |
      +------------+-------+-------+-------+-------+-------+-------+
      | A1         | +0.00 | +0.00 | -0.11 | -0.01 | 100.8 |  99.7 |
      | A2         | +0.01 | +0.02 | -0.01 | +0.01 | 100.2 | 100.1 |
      +------------+-------+-------+-------+-------+-------+-------+

[kslu-aom]

Please review this refactoring patch and let me know if anything can be improved. I have a follow up patch to clean up more variables, search state structs, and function signatures, but I think it is better to separate this unnesting change for code review. I will mark the PR as ready when speed 0 tests are finished.

[yeqing-wu]

@kslu-aom, thanks very much for helping refactor these code, it looks good to me. Just have one question that need your confirm.

[yeqing-wu]

The old code allocated int single_cwp_mask[MAX_CWP_NUM] = { 0 }; single_cwp_mask[0] = 1; on the stack and passed that. The new code passes NULL.

Would this cause a problem? Please check.

[kslu-aom]

This is a minor cleanup for handle_single_inter_prediction. Since CWP and JMVD are irrelevant to the single prediction case, the mask and cost are removed from here.