docs: add the changelog lines for three fixes already on main - #346

Summary

Adds three lines to the Unreleased section of CHANGELOG.md for changes that #327 put on main and that have no release note yet:

• a var array with a non-empty initializer declared inside an if block is initialized when the declaration is first reached (Known issue: a var array with an initializer declared inside an if block stays empty #333);
• strategy.risk.max_drawdown() and strategy.risk.max_intraday_loss() called with type = strategy.percent_of_equity, or with that constant held in a variable, apply the percentage limit like the positional form (Known issue: direction restrictions and percentage limits in strategy.risk calls #325, Known issue: some supplied arguments are ignored or shifted in v1.10.0 #326);
• max_bars_back(series, DEPTH) with a const int depth keeps that depth (Known issue: some supplied arguments are ignored or shifted in v1.10.0 #326).

The lines sit after the strategy.risk.allow_entry_in line of #292, in the style of their neighbours (issue and pull request numbers at the end, "Regenerate C++ and relink strategies." because the emitted C++ changes for these forms).

How the lines were checked

Each form was run next to its control on BTCUSDT 1h bars (2024-01-01 to 2024-03-01): the script was translated, built and run, and its trades were joined with the trades of the control by entry time and quantity. The control is the same script with the positional argument, the literal depth, or the var declared at the top level, so the two scripts differ in that one call or declaration only.

• On the released 1.10.0 (this package with its paired engine) every form differs from its control: the var array case places no order (the array stays empty); the keyword and variable risk forms place 1 entry (max_drawdown) and 62 entries (max_intraday_loss) against 144; the const depth case differs on 10 of 36 entries.
• On main (e32e9e4 with engine 06eb4962) every form equals its control: 36 of 36 entries for the var array case, 144 of 144 for each of the four risk forms, and 36 of 36 for the depth case.
• The commit that made them equal is Codegen: keyword binding, request contexts, array and numeric semantics, drawing collections, diagnostics #327: on its parent commit (Codegen train: timestamp date strings with a UTC or GMT zone word #330) every form still differs, on Codegen: keyword binding, request contexts, array and numeric semantics, drawing collections, diagnostics #327 every form equals its control.
• The repository's own test for the var array case is tests/test_conditional_var_array_first_reach.py.

What is not covered

• Documentation only: CHANGELOG.md is the only file changed, and no test of the repository reads it.
• The percentage forms are compared with the positional form of the same call, not with another tool. strategy.risk.max_drawdown() keeps its existing warning about partial runtime support.
• One symbol, one hourly window and one bar rhythm per script.

🤖 Generated with Claude Code