fix(ClusterStatistic): correct max value calculations in visible-proportion mode

[AlbertoAmadorBelchistim]

Summary

Two copy-paste bugs in CreateMaxValues produce incorrect color intensity on the cluster statistic table when the user enables
Visible Proportion. The non-VisibleProportion branch of the same method handles both cases correctly, which is what made these stand out.

Bugs fixed

1. maxBid reads from candle.Ask

// CreateMaxValues, VisibleProportion branch
maxAsk = Math.Max(candle.Ask, maxAsk);
maxBid = Math.Max(candle.Ask, maxBid);   // should be candle.Bid

The non-VisibleProportion path (_maxBid in OnCalculate) reads candle.Bid correctly. With this typo, the Bid row's rate is
normalised against Ask volumes, so on instruments with imbalanced Ask/Bid distribution the Bid column shows the wrong gradient.

2. maxDelta does not use Math.Abs

// CreateMaxValues, VisibleProportion branch
var maxDelta = 0m;
...
maxDelta = Math.Max(candle.Delta, maxDelta);   // signed

The non-VisibleProportion path uses Math.Max(Math.Abs(candle.Delta), _maxDelta), and GetRate(DataType.Delta, …) feeds Math.Abs(candle.Delta) as the value to normalise. When every visible bar has a negative delta, the signed Math.Max keeps maxDelta at its 0m initialiser; GetRate then hits the maxValue == 0 fallback (return 10) and flattens the gradient on the Delta row.

Repro

1. Open Cluster Statistic on any chart and enable Visible Proportion.
2. Scroll the visible window to a stretch where every bar has negative delta → Delta row gradient becomes uniform (rate = 10).
3. On a market with strongly imbalanced Ask/Bid volumes, compare the Bid row gradient with the non-VisibleProportion mode → they no longer match.

Risk

Minimal. Both changes are localised to the VisibleProportion branch of CreateMaxValues and align it with the existing global-mode behaviour. No public API changes, no new allocations, no flavor impact.

Not in this PR

While reviewing the file I spotted a handful of additional findings (missing OnDispose, asymmetric i == 0 guard in the same loop, Duration formatting without InvariantCulture, stale per-session last-alert values when SessionCumMode changes, minor whitespace).
None of them affect correctness in the same direct way as the two fixes above and they touch unrelated code paths, so I'm keeping this PR focused and will open a follow-up PR for the rest.