feat: `mutate()` constructs intermediate data frames for each new variable

[krlmlr]

For #270.

@toppyy: I believe this is the only instance of rel_translate() that gets called without data . This will simplify your PR. Appreciate your thoughts.

[github-actions[bot]]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if f03e52e7a6ad5d09e03cc3f07029402d783032a7 is merged into main:

• :heavy_check_mark:001_tpch_01: 23.2ms -> 22.6ms [-6.69%, +1.46%]
• :rocket:001_tpch_02: 117ms -> 116ms [-2.1%, -0.18%]
• :exclamation::snail:001_tpch_03: 60.2ms -> 61.3ms [+0.27%, +3.51%]
• :heavy_check_mark:001_tpch_04: 22.9ms -> 22.5ms [-3.18%, +0.12%]
• :heavy_check_mark:001_tpch_05: 112ms -> 111ms [-1.46%, +1%]
• :heavy_check_mark:001_tpch_06: 14.2ms -> 14.3ms [-2%, +3.86%]
• :heavy_check_mark:001_tpch_07: 128ms -> 129ms [-0.55%, +1.58%]
• :heavy_check_mark:001_tpch_08: 146ms -> 147ms [-0.63%, +1.7%]
• :heavy_check_mark:001_tpch_09: 120ms -> 120ms [-1.27%, +1.01%]
• :heavy_check_mark:001_tpch_10: 69ms -> 69.2ms [-0.83%, +1.29%]
• :heavy_check_mark:001_tpch_11: 58.7ms -> 57.8ms [-3.46%, +0.35%]
• :heavy_check_mark:001_tpch_12: 27.1ms -> 27.1ms [-2.38%, +1.73%]
• :heavy_check_mark:001_tpch_13: 18.8ms -> 18.8ms [-2.69%, +3.07%]
• :heavy_check_mark:001_tpch_14: 20.2ms -> 20.1ms [-3.48%, +2.74%]
• :heavy_check_mark:001_tpch_15: 57.9ms -> 57.9ms [-2.11%, +1.99%]
• :heavy_check_mark:001_tpch_16: 60.2ms -> 60ms [-2.08%, +1.14%]
• :heavy_check_mark:001_tpch_17: 25.5ms -> 25.5ms [-1.22%, +1.57%]
• :heavy_check_mark:001_tpch_18: 21.8ms -> 21.9ms [-1.82%, +2.67%]
• :heavy_check_mark:001_tpch_19: 119ms -> 119ms [-1.76%, +0.79%]
• :heavy_check_mark:001_tpch_20: 71.4ms -> 71.6ms [-0.91%, +1.68%]
• :heavy_check_mark:001_tpch_21: 161ms -> 160ms [-1.63%, +1%]
• :heavy_check_mark:001_tpch_22: 115ms -> 116ms [-0.25%, +2.77%]
• :heavy_check_mark:010_tpch_01: 78.8ms -> 79.1ms [-5.45%, +6.29%]
• :heavy_check_mark:010_tpch_02: 66.5ms -> 67.1ms [-2.33%, +4.17%]
• :heavy_check_mark:010_tpch_03: 63.1ms -> 61.2ms [-7.88%, +1.79%]
• :rocket:010_tpch_04: 47.3ms -> 45.7ms [-6.1%, -0.68%]
• :heavy_check_mark:010_tpch_05: 88.6ms -> 89.9ms [-2.16%, +5.24%]
• :heavy_check_mark:010_tpch_06: 33.3ms -> 34.3ms [-4.27%, +10.54%]
• :heavy_check_mark:010_tpch_07: 108ms -> 106ms [-6.01%, +2.15%]
• :heavy_check_mark:010_tpch_08: 124ms -> 126ms [-0.15%, +3.61%]
• :heavy_check_mark:010_tpch_09: 111ms -> 112ms [-3.1%, +4.84%]
• :heavy_check_mark:010_tpch_10: 78.8ms -> 78.2ms [-3.9%, +2.55%]
• :heavy_check_mark:010_tpch_11: 36.9ms -> 36.1ms [-8.84%, +4.22%]
• :heavy_check_mark:010_tpch_12: 58.2ms -> 60.5ms [-1.42%, +9.36%]
• :heavy_check_mark:010_tpch_13: 49.1ms -> 48.9ms [-2.96%, +2.16%]
• :heavy_check_mark:010_tpch_14: 38.8ms -> 38.8ms [-2.55%, +2.64%]
• :heavy_check_mark:010_tpch_15: 54.9ms -> 54.9ms [-6.81%, +6.64%]
• :heavy_check_mark:010_tpch_16: 41.3ms -> 41.5ms [-2.49%, +3.28%]
• :heavy_check_mark:010_tpch_17: 53.6ms -> 52.3ms [-7.82%, +3.14%]
• :heavy_check_mark:010_tpch_18: 53.6ms -> 51.7ms [-9.4%, +2.4%]
• :heavy_check_mark:010_tpch_19: 116ms -> 114ms [-4.02%, +1.13%]
• :heavy_check_mark:010_tpch_20: 67.9ms -> 67.3ms [-4.08%, +2.36%]
• :heavy_check_mark:010_tpch_21: 391ms -> 392ms [-1.32%, +1.79%]
• :heavy_check_mark:010_tpch_22: 73.2ms -> 71.5ms [-6.3%, +1.75%]
• :heavy_check_mark:100_tpch_01: 1.14s -> 1.11s [-5.43%, +0.21%]
• :heavy_check_mark:100_tpch_02: 114ms -> 113ms [-3.51%, +2.25%]
• :heavy_check_mark:100_tpch_03: 1.02s -> 1.02s [-2.85%, +3.11%]
• :heavy_check_mark:100_tpch_04: 1s -> 988ms [-5.1%, +1.69%]
• :heavy_check_mark:100_tpch_05: 1.1s -> 1.08s [-6.17%, +1.85%]
• :heavy_check_mark:100_tpch_06: 939ms -> 936ms [-2.63%, +1.86%]
• :heavy_check_mark:100_tpch_07: 1.06s -> 1.06s [-2.31%, +3.69%]
• :heavy_check_mark:100_tpch_08: 1.09s -> 1.09s [-3.62%, +4.86%]
• :heavy_check_mark:100_tpch_09: 1.16s -> 1.17s [-2.18%, +3.98%]
• :heavy_check_mark:100_tpch_10: 1.05s -> 1.05s [-2.58%, +1.55%]
• :heavy_check_mark:100_tpch_11: 83.2ms -> 76.3ms [-20.86%, +4.25%]
• :heavy_check_mark:100_tpch_12: 1.03s -> 1.02s [-2.73%, +0.9%]
• :heavy_check_mark:100_tpch_13: 299ms -> 295ms [-9.54%, +6.95%]
• :heavy_check_mark:100_tpch_14: 954ms -> 955ms [-1.03%, +1.16%]
• :heavy_check_mark:100_tpch_15: 1.03s -> 1.05s [-0.02%, +3.92%]
• :rocket:100_tpch_16: 119ms -> 114ms [-7.48%, -1.36%]
• :heavy_check_mark:100_tpch_17: 1.01s -> 1.01s [-2.47%, +3.19%]
• :heavy_check_mark:100_tpch_18: 1.03s -> 1.03s [-2.57%, +3.31%]
• :heavy_check_mark:100_tpch_19: 1.12s -> 1.12s [-1.6%, +1.51%]
• :heavy_check_mark:100_tpch_20: 1.01s -> 1.02s [-0.28%, +3.97%]
• :heavy_check_mark:100_tpch_21: 2.08s -> 2.08s [-2.78%, +2.72%]
• :rocket:100_tpch_22: 173ms -> 159ms [-13.34%, -3.26%]

Further explanation regarding interpretation and methodology can be found in the documentation.

[toppyy]

Thanks.

So if I'm understanding correctly, now that there are intermediate data frames we could pass them as arguments like this rel_translate(quo, data = current_data, etc..)? This would allow us to 1) make comparison expressions also within mutate() and 2) skip checking if datais missing as rel_translate does not get called without it.

A sidenote: I explored the above approach and found that an unnecessary transformation within to_duckdb_expr leads to an error if the expr has an alias. I removed the transformation in the PR.