Fix 'fast reduce' variable incorrectly is set to 0 as its default

[SSoelvsten]

The default constructor for exec_policy::nested::fast_reduce passes a float 0.05 directly to the unsigned char _value. This immediately converts it to the value of 0 rather than running it through the from_float function as intended.

Based on various experiments, we can also conclude that:

• The epsilon value might need to be different for the inner and the outer Reduce?
• The value for fast reduce seems to better as -1.0 (turned off) or possibly 0.02.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 97.306%. Comparing base (5ac21d0) to head (a59ca92).
Report is 3 commits behind head on main.

❗ Current head a59ca92 differs from pull request most recent head 29a32ac. Consider uploading reports for the commit 29a32ac to get more accurate results

Additional details and impacted files

@@              Coverage Diff              @@
##              main      #640       +/-   ##
=============================================
- Coverage   97.306%   97.306%   -0.000%     
=============================================
  Files           90        90               
  Lines         6534      6533        -1     
=============================================
- Hits          6358      6357        -1     
  Misses         176       176               

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

🔴 Regression Test (QBF 'ep/8x8_7_e-8-1_p-3-4_bwnib')

'origin/exec_policy/bug/fast_reduce-always-zero' is a change in performance of -0.46% (stdev: 0.24%).

...	origin/main	origin/exec_policy/bug/fast_reduce-always-zero
Mean	11391.67	11444.33
Standard Deviation	27.21	11.50

Number of samples: 3

[github-actions]

🔴 Regression Test (QBF 'domineering/5x5_13_bwnib')

'origin/exec_policy/bug/fast_reduce-always-zero' is a change in performance of -51.07% (stdev: 0.48%).

...	origin/main	origin/exec_policy/bug/fast_reduce-always-zero
Mean	22767.00	34394.33
Standard Deviation	29.51	166.78

Number of samples: 3

[github-actions]

🟡 Regression Test (QBF 'breakthrough/3x4_19_bwnib')

'origin/exec_policy/bug/fast_reduce-always-zero' is a change in performance of -0.57% (stdev: 0.68%).

...	origin/main	origin/exec_policy/bug/fast_reduce-always-zero
Mean	24472.33	24611.67
Standard Deviation	167.05	42.44

Number of samples: 3

[github-actions]

🔴 Regression Test (QBF 'breakthrough/3x5_11_bwnib')

'origin/exec_policy/bug/fast_reduce-always-zero' is a change in performance of -13.91% (stdev: 0.26%).

...	origin/main	origin/exec_policy/bug/fast_reduce-always-zero
Mean	7349.25	8371.75
Standard Deviation	19.19	12.42

Number of samples: 4

[github-actions]

🟢 Regression Test (QBF 'connect4/4x4_9_connect3_bwnib')

'origin/exec_policy/bug/fast_reduce-always-zero' is a change in performance of 2.43% (stdev: 9.51%).

...	origin/main	origin/exec_policy/bug/fast_reduce-always-zero
Mean	1219.80	1190.10
Standard Deviation	115.98	17.00

Number of samples: 10

[SSoelvsten]

Based on these experiments, this optimisation should just be removed. This is going to be done on a separate branch and pull request instead.