Emit GeneratedStrategy on pineforge::source::PineStrategyHost (R4-C source-layer cut, paired with engine)

[luisleo526]

Paired codegen half of the engine R4-C source-layer cut (roadmap slice B). Generated behaviour is unchanged: this PR only re-targets the generated class at the explicit source host.

Landing 1 (this head)

• #include <pineforge/source/pine_strategy_host.hpp> replaces the engine include.
• class GeneratedStrategy : public pineforge::source::PineStrategyHost.
• The bar callback is emitted as on_source_bar(const Bar&) override (the host's on_bar is final and forwards).
• Constructor writes, set_strategy_override, the cap/priority attach block and the run/stream emission are unchanged here (landing 4 moves them onto the declared PineStrategyConfig/adapter surface).
• Golden diff = exactly three substitutions; support_checker, signatures, official-surface tests unchanged.

Verification: full pytest with the compile-backed suites pointed at the engine L1 include tree: 2813 passed, 2 skipped.

Still to come on this branch

Landing 4 (config initializer + adapter override + risk setters) after the engine transfer lands; then ONE composite Cloud sweep (engine head + this head) with zero-diff readback, snapshot and the actual pr-gate before this PR is marked ready. No R4 credit is claimed by this slice.

🤖 Generated with Claude Code

https://claude.ai/code/session_016iF71UAEuQ7F6eqWgecAHm

[luisleo526]

L4 landed: head 8a8f330. The generated constructor now builds one pineforge::source::PineStrategyConfig cfg{} (same 13 facts, same order) and makes exactly one configure_pine_strategy(cfg) call before any metadata; set_strategy_override fills a source::StrategyOverrides (10 keys, same conversions) and calls the host entry once; strategy.risk.* statements emit the six explicit source-host setters (max_intraday_filled_orders → set_pine_risk_max_intraday_filled_orders((int)expr), the old integer-assignment attach). run_backtest/stream emission, visit_call.py, support_checker, signatures, official-surface tests unchanged. Golden: +40/−27, all inside the constructor/override/risk blocks.

pytest: 1926 passed offline; 2816 passed / 2 skipped with the compile-backed suites against the engine source-layer headers (incl. 314 corpus compiles). Fable final review and the composite Cloud sweep (engine 6e9b908 + this head) follow before ready.

[luisleo526]

L5 landed: head 763d2eb (tests only, +123 lines). Pins the emitted Pine direction convention (.short → set_pine_risk_direction(-1), .long → 1, .all → 0) and the percent-flag argument (true only for strategy.percent_of_equity), plus two compile-backed runtime probes proving a short-only policy drops a long entry and that percent-then-cash keeps the sticky percent interpretation — both against the engine L5 head 2c964db (the engine-side mapping/sticky-flag repairs from the final review). Full pytest with compile-backed suites: 2820 passed / 2 skipped.

[luisleo526]

Engine half merged as ab9714b (PR #253). Final pair evidence: composite sweep exp-native-r4-c-1938fe0-20260914 4190/4190 zero trade/grade/metadata/coverage differences, actual pr-gate FAIL target.not-positive (expected neutral), baseline stable; this head 763d2eb compile-backed against the engine head: 2820 passed. Marking ready and squash-merging.