bug in refactored compute_grid_info()

[simonpcouch]

After merging #957, we saw the following failures on extratests over the weekend:

══ Failed ══════════════════════════════════════════════════════════════════════
── 1. Failure ('test-mars-tuning.R:63:3'): tuning for mars() -- submodels *and* 
mars_metrics$.config (`actual`) not equal to paste0("Preprocessor1_Model", 1:7) (`expected`).

    actual                 | expected                  
[1] "Preprocessor1_Model1" | "Preprocessor1_Model1" [1]
[2] "Preprocessor1_Model1" - "Preprocessor1_Model2" [2]
[3] "Preprocessor1_Model2" - "Preprocessor1_Model3" [3]
[4] "Preprocessor1_Model2" - "Preprocessor1_Model4" [4]
[5] "Preprocessor1_Model3" - "Preprocessor1_Model5" [5]
[6] "Preprocessor1_Model4" - "Preprocessor1_Model6" [6]
[7] "Preprocessor1_Model5" - "Preprocessor1_Model7" [7]

── 2. Failure ('test-mars-tuning.R:68:3'): tuning for mars() -- submodels *and* 
count(mars_preds, .row, .config)$n (`actual`) not equal to rep(1L, 5537) (`expected`).

       actual | expected       
   [4] 1      | 1        [1]   
   [5] 1      | 1        [2]   
   [6] 2      - 1        [3]   
   [7] 2      - 1        [4]   

Some additional trickiness here in grids that are partially regular and partially not so.