Fix indicator, cache, and trading analysis regressions

[sdcoffey]

Indicator composition could seed averages with warm-up placeholders, retain stale values after candle edits, and accumulate quadratic EMA/MMA memory usage. Several indicator, crossover, and trade-analysis edge cases also produced incorrect results. This fixes the audit findings while retaining Go 1.21 and the existing module dependencies.

Changes

• Calculate EMA/MMA histories iteratively with cached state rounded to 256-bit precision. Add FirstValidIndex metadata and propagate cache invalidation through built-in indicator dependencies.
• Correct crossover events across equality plateaus, chronological maximum drawdown, CCI typical-price deviation, Aroon evaluation order and tied extremes, Bollinger warm-up, and Wilder RSI initialization.
• Correct directional profit reporting, fractional-period returns, buy-and-hold candle horizons, and the strategy warm-up boundary. Validate malformed and reversed time periods without panicking.
• Move test helpers out of production builds, make fixtures deterministic, repair and test the examples, pin developer tools, and add Go 1.27 and lint checks to CI.

Calculation fixes can change existing indicator values, trading signals, and analysis results. EMA/MMA now retain roughly 77 significant decimal digits rather than growing precision indefinitely. The README and changelog document these conventions, cache reset requirements, and return-calculation assumptions.

Related reports: #18, #25, #53, #49, #50, and #13.

Validation

• Race-enabled tests pass on Go 1.21.13, 1.26.3, and 1.27.1; vet passes on the minimum and latest versions, and make lint passes with pinned Staticcheck.
• Go 1.21 tests on the current macOS host used -ldflags=-linkmode=external to avoid the old internal linker's missing LC_UUID issue.
• Parser fuzzing passed for 10 seconds (31,676 executions).
• Regression tests cover composition, candle mutation and reindexing, precision and numeric scale, indicators, crossovers, trade analysis, parser validation, and strategy boundaries.
• EMA/MMA history benchmarks at 400, 800, and 1,600 samples show approximately linear allocation growth. A local 1,600-sample EMA reproduction reduced retained heap from about 82.5 MiB to 0.26 MiB.
• git diff --check passes; go.mod and go.sum are unchanged.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-04T22:56:55.454522Z	8fdcfa4	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8fdcfa4c78

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".