Improve compare.py output to use min times and better column titles

[alamb]

Which issue does this PR close?

Part of #6127

Rationale for this change

I want to be able to easily understand what is being compared and reported. Currently the reporting from compare.py is somewhat opaque:

1. It averages benchmark run timings (which increases the variance)
2. The header columns are non sensical

What changes are included in this PR?

1. Use min of the iteration times to report the benchmark results -- I think this minimizes run to run variance and is the fairest way to asses performance
2. Use the filename as the column header

Are these changes tested?

No, it is a tool only change

Are there any user-facing changes?

Not really

Example output Before this PR

(note the bad headers)

python3 compare.py results/alamb_bench/tpch.json results/alamb_bench/tpch_mem.json
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃ /Users/alam… ┃           -o ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1     │     794.64ms │     559.37ms │ +1.42x faster │
│ QQuery 2     │     239.12ms │     171.29ms │ +1.40x faster │
│ QQuery 3     │     254.99ms │      99.33ms │ +2.57x faster │
│ QQuery 4     │     147.85ms │      66.65ms │ +2.22x faster │
│ QQuery 5     │     322.94ms │     258.98ms │ +1.25x faster │
│ QQuery 6     │     130.89ms │      24.24ms │ +5.40x faster │
│ QQuery 7     │     629.42ms │     645.68ms │     no change │
│ QQuery 8     │     375.57ms │     140.26ms │ +2.68x faster │
│ QQuery 9     │     597.65ms │     334.59ms │ +1.79x faster │
│ QQuery 10    │     397.16ms │     197.59ms │ +2.01x faster │
│ QQuery 11    │     182.41ms │     167.33ms │ +1.09x faster │
│ QQuery 12    │     228.44ms │      99.10ms │ +2.31x faster │
│ QQuery 13    │     736.21ms │     438.47ms │ +1.68x faster │
│ QQuery 14    │     195.68ms │      31.99ms │ +6.12x faster │
│ QQuery 15    │     175.08ms │      57.62ms │ +3.04x faster │
│ QQuery 16    │     175.67ms │     144.90ms │ +1.21x faster │
│ QQuery 17    │    1953.59ms │    1925.39ms │     no change │
│ QQuery 18    │    2133.06ms │    1968.85ms │ +1.08x faster │
│ QQuery 19    │     371.82ms │     107.44ms │ +3.46x faster │
│ QQuery 20    │     680.02ms │     567.75ms │ +1.20x faster │
│ QQuery 21    │     987.36ms │     857.87ms │ +1.15x faster │
│ QQuery 22    │     105.31ms │      77.25ms │ +1.36x faster │
└──────────────┴──────────────┴──────────────┴───────────────┘

Output after this PR

python3 compare.py results/alamb_bench/tpch.json results/alamb_bench/tpch_mem.json
┏━━━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Query        ┃      tpch ┃  tpch_mem ┃        Change ┃
┡━━━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ QQuery 1     │  762.35ms │  548.16ms │ +1.39x faster │
│ QQuery 2     │  210.23ms │  164.26ms │ +1.28x faster │
│ QQuery 3     │  239.37ms │   96.47ms │ +2.48x faster │
│ QQuery 4     │  147.01ms │   59.81ms │ +2.46x faster │
│ QQuery 5     │  317.68ms │  251.01ms │ +1.27x faster │
│ QQuery 6     │  129.40ms │   22.86ms │ +5.66x faster │
│ QQuery 7     │  603.09ms │  629.24ms │     no change │
│ QQuery 8     │  359.94ms │  138.52ms │ +2.60x faster │
│ QQuery 9     │  578.71ms │  329.58ms │ +1.76x faster │
│ QQuery 10    │  378.17ms │  188.56ms │ +2.01x faster │
│ QQuery 11    │  180.20ms │  159.49ms │ +1.13x faster │
│ QQuery 12    │  219.37ms │   94.54ms │ +2.32x faster │
│ QQuery 13    │  732.80ms │  425.82ms │ +1.72x faster │
│ QQuery 14    │  190.55ms │   30.99ms │ +6.15x faster │
│ QQuery 15    │  173.51ms │   54.58ms │ +3.18x faster │
│ QQuery 16    │  163.33ms │  130.02ms │ +1.26x faster │
│ QQuery 17    │ 1866.25ms │ 1855.35ms │     no change │
│ QQuery 18    │ 2063.31ms │ 1916.26ms │ +1.08x faster │
│ QQuery 19    │  363.00ms │  102.86ms │ +3.53x faster │
│ QQuery 20    │  639.92ms │  530.65ms │ +1.21x faster │
│ QQuery 21    │  946.66ms │  815.83ms │ +1.16x faster │
│ QQuery 22    │  103.37ms │   73.07ms │ +1.41x faster │
└──────────────┴───────────┴───────────┴───────────────┘

[comphead]

lgtm, but imho, can min be deceptive comparing to avg/mean/median? 🤔

[alamb]

I agree it is misleading -- in terms of measuring a change between datafusion versions, I think min will give us the least variance between runs and represents best case performance.

However, it doesn't really give a sense for how much variation is across runs 🤔