perf: Improve decimal addition and subtraction when scale is equal

[AdamGS]

Which issue does this PR close?

• Closes #NNN.

Rationale for this change

For cases where adding or subtracting decimal values that have the same scale, we end up doing a bunch of extra checked mul operations, and turns out if we don't it improves performance significantly.

What changes are included in this PR?

1. Specialize add/sub for decimals when they have the same scale.
2. Benchmark to measure the impact on all decimals types.

Are these changes tested?

Existing tests, 1 additional test I added to i256 which was my original target.

Are there any user-facing changes?

None

[AdamGS]

On my laptop, from 40% improvement for decimal32 and up to 80% on decimal 256:

decimal32_equal_scale/add
                        time:   [630.04 ns 630.58 ns 631.29 ns]
                        thrpt:  [1.6221 Gelem/s 1.6239 Gelem/s 1.6253 Gelem/s]
                 change:
                        time:   [−43.418% −43.295% −43.187%] (p = 0.00 < 0.05)
                        thrpt:  [+76.016% +76.352% +76.736%]
                        Performance has improved.
Found 9 outliers among 100 measurements (9.00%)
  2 (2.00%) low mild
  5 (5.00%) high mild
  2 (2.00%) high severe
decimal32_equal_scale/sub
                        time:   [630.33 ns 630.63 ns 630.98 ns]
                        thrpt:  [1.6229 Gelem/s 1.6238 Gelem/s 1.6245 Gelem/s]
                 change:
                        time:   [−43.643% −43.516% −43.400%] (p = 0.00 < 0.05)
                        thrpt:  [+76.678% +77.042% +77.441%]
                        Performance has improved.
Found 16 outliers among 100 measurements (16.00%)
  13 (13.00%) high mild
  3 (3.00%) high severe

decimal32_different_scale/add
                        time:   [1.1945 µs 1.1951 µs 1.1957 µs]
                        thrpt:  [856.40 Melem/s 856.84 Melem/s 857.23 Melem/s]
                 change:
                        time:   [+0.1629% +0.8429% +1.5190%] (p = 0.02 < 0.05)
                        thrpt:  [−1.4963% −0.8359% −0.1626%]
                        Change within noise threshold.
Found 9 outliers among 100 measurements (9.00%)
  2 (2.00%) low severe
  1 (1.00%) low mild
  1 (1.00%) high mild
  5 (5.00%) high severe
decimal32_different_scale/sub
                        time:   [1.1870 µs 1.1886 µs 1.1900 µs]
                        thrpt:  [860.50 Melem/s 861.53 Melem/s 862.69 Melem/s]
                 change:
                        time:   [+1.5117% +1.9507% +2.3788%] (p = 0.00 < 0.05)
                        thrpt:  [−2.3235% −1.9134% −1.4892%]
                        Performance has regressed.
Found 9 outliers among 100 measurements (9.00%)
  1 (1.00%) low severe
  7 (7.00%) high mild
  1 (1.00%) high severe

decimal64_equal_scale/add
                        time:   [639.33 ns 642.21 ns 645.03 ns]
                        thrpt:  [1.5875 Gelem/s 1.5945 Gelem/s 1.6017 Gelem/s]
                 change:
                        time:   [−45.814% −45.629% −45.466%] (p = 0.00 < 0.05)
                        thrpt:  [+83.371% +83.923% +84.549%]
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high mild
decimal64_equal_scale/sub
                        time:   [629.79 ns 630.77 ns 632.16 ns]
                        thrpt:  [1.6198 Gelem/s 1.6234 Gelem/s 1.6259 Gelem/s]
                 change:
                        time:   [−46.014% −45.884% −45.727%] (p = 0.00 < 0.05)
                        thrpt:  [+84.253% +84.790% +85.234%]
                        Performance has improved.
Found 6 outliers among 100 measurements (6.00%)
  4 (4.00%) high mild
  2 (2.00%) high severe

decimal64_different_scale/add
                        time:   [1.1947 µs 1.1960 µs 1.1978 µs]
                        thrpt:  [854.92 Melem/s 856.21 Melem/s 857.12 Melem/s]
                 change:
                        time:   [+2.3087% +2.6107% +2.9771%] (p = 0.00 < 0.05)
                        thrpt:  [−2.8910% −2.5443% −2.2566%]
                        Performance has regressed.
Found 9 outliers among 100 measurements (9.00%)
  1 (1.00%) low mild
  2 (2.00%) high mild
  6 (6.00%) high severe
decimal64_different_scale/sub
                        time:   [1.2057 µs 1.2108 µs 1.2157 µs]
                        thrpt:  [842.30 Melem/s 845.74 Melem/s 849.27 Melem/s]
                 change:
                        time:   [+1.6586% +1.9491% +2.2576%] (p = 0.00 < 0.05)
                        thrpt:  [−2.2077% −1.9119% −1.6315%]
                        Performance has regressed.
Found 14 outliers among 100 measurements (14.00%)
  11 (11.00%) high mild
  3 (3.00%) high severe

decimal128_equal_scale/add
                        time:   [763.21 ns 765.54 ns 767.91 ns]
                        thrpt:  [1.3335 Gelem/s 1.3376 Gelem/s 1.3417 Gelem/s]
                 change:
                        time:   [−72.061% −71.976% −71.891%] (p = 0.00 < 0.05)
                        thrpt:  [+255.76% +256.84% +257.92%]
                        Performance has improved.
Found 15 outliers among 100 measurements (15.00%)
  9 (9.00%) low mild
  5 (5.00%) high mild
  1 (1.00%) high severe
decimal128_equal_scale/sub
                        time:   [765.55 ns 767.56 ns 769.39 ns]
                        thrpt:  [1.3309 Gelem/s 1.3341 Gelem/s 1.3376 Gelem/s]
                 change:
                        time:   [−72.614% −72.438% −72.274%] (p = 0.00 < 0.05)
                        thrpt:  [+260.68% +262.82% +265.15%]
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high mild

decimal128_different_scale/add
                        time:   [2.7359 µs 2.7407 µs 2.7465 µs]
                        thrpt:  [372.84 Melem/s 373.63 Melem/s 374.28 Melem/s]
                 change:
                        time:   [−1.5850% −1.3380% −1.0742%] (p = 0.00 < 0.05)
                        thrpt:  [+1.0859% +1.3561% +1.6106%]
                        Performance has improved.
Found 10 outliers among 100 measurements (10.00%)
  1 (1.00%) low mild
  2 (2.00%) high mild
  7 (7.00%) high severe
decimal128_different_scale/sub
                        time:   [2.7348 µs 2.7378 µs 2.7412 µs]
                        thrpt:  [373.55 Melem/s 374.03 Melem/s 374.44 Melem/s]
                 change:
                        time:   [−1.8846% −1.6337% −1.3612%] (p = 0.00 < 0.05)
                        thrpt:  [+1.3800% +1.6608% +1.9208%]
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  4 (4.00%) low mild
  3 (3.00%) high severe

decimal256_equal_scale/add
                        time:   [1.5181 µs 1.5189 µs 1.5198 µs]
                        thrpt:  [673.79 Melem/s 674.15 Melem/s 674.52 Melem/s]
                 change:
                        time:   [−86.348% −86.331% −86.315%] (p = 0.00 < 0.05)
                        thrpt:  [+630.71% +631.57% +632.50%]
                        Performance has improved.
decimal256_equal_scale/sub
                        time:   [1.4986 µs 1.5010 µs 1.5041 µs]
                        thrpt:  [680.79 Melem/s 682.20 Melem/s 683.31 Melem/s]
                 change:
                        time:   [−86.498% −86.478% −86.457%] (p = 0.00 < 0.05)
                        thrpt:  [+638.38% +639.55% +640.61%]
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high severe

decimal256_different_scale/add
                        time:   [11.071 µs 11.079 µs 11.090 µs]
                        thrpt:  [92.337 Melem/s 92.429 Melem/s 92.490 Melem/s]
                 change:
                        time:   [−0.9168% −0.6735% −0.4525%] (p = 0.00 < 0.05)
                        thrpt:  [+0.4545% +0.6781% +0.9253%]
                        Change within noise threshold.
Found 4 outliers among 100 measurements (4.00%)
  1 (1.00%) low severe
  2 (2.00%) low mild
  1 (1.00%) high severe
decimal256_different_scale/sub
                        time:   [11.095 µs 11.101 µs 11.109 µs]
                        thrpt:  [92.178 Melem/s 92.247 Melem/s 92.291 Melem/s]
                 change:
                        time:   [−0.5163% −0.3530% −0.2031%] (p = 0.00 < 0.05)
                        thrpt:  [+0.2035% +0.3542% +0.5190%]
                        Change within noise threshold.
Found 6 outliers among 100 measurements (6.00%)
  1 (1.00%) low severe
  3 (3.00%) high mild
  2 (2.00%) high severe

[Jefffrey]

thanks @AdamGS