feat: add EwmaChangeDetector, an EWMA control chart for streams

[alxkm]

Adds an EWMA control chart, the exponentially weighted counterpart of the CUSUM chart that went in with #7593: it smooths the stream with an average seeded at the expected level and raises an alarm as soon as the smoothed value leaves a band around that level.

z     <- z + alpha * (x - target)
limit  = width * sigma * sqrt( alpha / (2 - alpha) * (1 - (1 - alpha)^(2n)) )

The band is the exact standard deviation of z under the null hypothesis, scaled by the desired width in sigmas, so it starts narrow and widens towards its asymptote width * sigma * sqrt(alpha / (2 - alpha)). Using that asymptote from the first sample onwards -- the usual shortcut -- leaves the chart far too tolerant while the average is still warming up, whereas the exact limit keeps the false alarm rate steady from the very first sample. On an alarm the average is returned to the target, so the detector starts fresh on the next change instead of latching and reporting the same shift forever.

Where a CUSUM chart accumulates evidence without limit and so excels at small persistent shifts, an EWMA chart looks at a decaying window of the recent past: alpha around 0.1 to 0.3 is a good compromise, larger values reacting faster to big jumps and smaller ones being more sensitive to slow drifts. Three sigmas is the customary width, and both are the defaults.

ExponentialMovingAverage carries the recurrence and is the reusable half of this: alpha can be given directly, as a span (2 / (span + 1), matching the responsiveness of a simple moving average of that many samples) or as a half-life (1 - exp(-ln2 / halfLife)), and the class tracks the exponentially weighted variance of the same stream in the same O(1) state. It seeds itself with the first sample rather than with zero, which avoids the warm-up bias a zero seed introduces, or with an explicit resting level, which is what a control chart wants.

The ShiftSignal verdict enum, until now nested inside CusumDetector, becomes a top-level enum of the package so that both detectors report the same type; CusumDetector and its test are updated to it, and nothing else in the repository referred to the nested enum.

Each sample costs O(1) time and neither class allocates anything after construction.

EwmaChangeDetectorTest covers 17 cases and ExponentialMovingAverageTest 31; CusumDetectorTest keeps its 16. Among them: the control limit is checked against its closed form and against its asymptote, upward and downward shifts are found both in a clean step and buried in noise, in-control Gaussian noise rarely trips the band, an alarm returns the statistic to the target, a stream sitting on target never alarms, the mean and variance recurrences are checked against direct computation, alpha == 1 keeps only the latest sample, and on stationary noise the estimates sit close to the true moments.

Checklist

• I have read CONTRIBUTING.md.
• This pull request is all my own work -- I have not plagiarized it.
• All filenames are in PascalCase.
• All functions and variable names follow Java naming conventions.
• All new algorithms have a URL in their comments that points to Wikipedia or other similar explanations.
• All new algorithms include a corresponding test class that validates their functionality.
• All new code is formatted with clang-format -i --style=file path/to/your/file.java

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 80.99%. Comparing base (d93d0ea) to head (7266171).

Additional details and impacted files

@@             Coverage Diff              @@
##             master    #7600      +/-   ##
============================================
+ Coverage     80.92%   80.99%   +0.07%     
- Complexity     7680     7737      +57     
============================================
  Files           822      825       +3     
  Lines         24490    24594     +104     
  Branches       4800     4810      +10     
============================================
+ Hits          19818    19920     +102     
- Misses         3906     3907       +1     
- Partials        766      767       +1     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.