perf: reduce linear regression channel allocations

[leo1987820]

Summary

Refs #2.

This PR optimizes linear_regression_channel without overlapping the existing performance submissions for dashboard grouping, stochastic/RSI/ATR/MACD/OBV, PnL chart rendering, or log buffering.

The previous implementation allocated x = list(range(lookback)) and a full residuals list on every call. This version:

• computes fixed sx and sxx with arithmetic-series formulas
• uses enumerate(y) instead of materializing x
• accumulates residual squared error directly instead of allocating residuals
• keeps the previous output behavior covered by regression tests

Validation

Because my local environment cannot currently clone this repository over GitHub TLS, I validated the exact touched module in a local minimal workspace.

Manual test invocation passed:

manual statistical_algos tests passed

Micro-benchmark over 2,000 calls on a 5,000-point price series:

lookback=20  old=0.019192s  new=0.012716s  speedup=1.51x
lookback=100 old=0.061628s  new=0.036901s  speedup=1.67x
lookback=500 old=0.284800s  new=0.188338s  speedup=1.51x

Notes

I intentionally avoided duplicating the existing open PRs and focused on a separate statistical strategy hot path.

[vercel]

@leo1987820 is attempting to deploy a commit to the dragoon4002's projects Team on Vercel.

A member of the Team first needs to authorize it.