feat: extend udf_overhead benchmark w/ sub-str example

[crepererum]

For #28.

Results

=== cost model ===                                                                                                                                                                                                 
y = cost_row * n_rows + cost_call * n_batches + cost_cached                                                                                                                                                        
                                                                                                                                                                                                                   
   cost_row: Cost per input row/cell.                                                                                                                                                                              
  cost_call: Cost per UDF call / record batch.                                                                                                                                                                     
cost_cached: Cost amortized by calling the UDF multiple times.                                                                                                                                                     
      score: How well the linear model fits (0=not, 1=perfect).                                                                                                                                                    
┌─────────┬────────┬──────────┬───────────┬─────────────┬───────┐                                                                                                                                                  
│ payload ┆ mode   ┆ cost_row ┆ cost_call ┆ cost_cached ┆ score │                                                                                                                                                  
╞═════════╪════════╪══════════╪═══════════╪═════════════╪═══════╡                                                                                                                                                  
│ addone  ┆ native ┆      110 ┆         0 ┆      15_142 ┆  1.00 │                                                                                                                                                  
│ addone  ┆ wasm   ┆      477 ┆   936_577 ┆     980_115 ┆  0.99 │                                                                                                                                                  
│ addone  ┆ python ┆    2_733 ┆   968_911 ┆     853_064 ┆  1.00 │                                                                                                                                                  
│ substr  ┆ native ┆    3_369 ┆         0 ┆           0 ┆  1.00 │                                                                                                                                                  
│ substr  ┆ wasm   ┆   12_390 ┆ 2_611_558 ┆     489_419 ┆  1.00 │                                                                                                                                                  
│ substr  ┆ python ┆   19_973 ┆ 2_964_281 ┆     657_907 ┆  1.00 │                                                                                                                                                  
└─────────┴────────┴──────────┴───────────┴─────────────┴───────┘    

=== delta rel2 ===
delta = y / max(x, 1)
┌─────────┬────────┬──────────┬──────────┬──────────────┬─────────────┐
│ payload ┆ mode   ┆ baseline ┆ cost_row ┆    cost_call ┆ cost_cached │
╞═════════╪════════╪══════════╪══════════╪══════════════╪═════════════╡
│ addone  ┆ wasm   ┆ native   ┆     4.34 ┆   936_577.00 ┆       64.73 │
│ addone  ┆ python ┆ native   ┆    24.85 ┆   968_911.00 ┆       56.34 │
│ addone  ┆ python ┆ wasm     ┆     5.73 ┆         1.03 ┆        0.87 │
│ substr  ┆ wasm   ┆ native   ┆     3.68 ┆ 2_611_558.00 ┆  489_419.00 │
│ substr  ┆ python ┆ native   ┆     5.93 ┆ 2_964_281.00 ┆  657_907.00 │
│ substr  ┆ python ┆ wasm     ┆     1.61 ┆         1.14 ┆        1.34 │
└─────────┴────────┴──────────┴──────────┴──────────────┴─────────────┘

The per-row overhead looks better if we pass larger data around, likely because the cost to manage python objects gets relatively lower. However I'm somewhat surprised that the per-call overhead for both Python and the plain WASM wrapper is soo large. Gonna need to check the reason for this in a follow-up.