feat(security): LTF input-passthrough when req >= input AND req < script_tf

[luisleo526]

Summary

• request.security_lower_tf previously rejected requests where requested TF was >= input TF, even when finer than script TF. Pine returns the real underlying input bars in that range; the engine now does too.
• New SecurityEvalState fields (lower_tf_use_input, lower_tf_input_aggregation_ratio, lower_tf_input_buffer) buffer raw input bars per state and dispatch one evaluate_security per resulting LTF bar at script-TF chunk completion. OHLCV roll-up when req > input.
• Validation in validate_security_timeframes accepts the new range (req >= input, req < script_tf, integer divisor of script TF, integer multiple of input TF) and continues to reject other cases with precise messages.
• Synthesis path (req < input_tf) and HTF path unchanged.

Test plan

• ctest: 22/22 pass (was 21; new test_security_lower_tf_input_passthrough covers raw passthrough, aggregated passthrough, req==script reject, synthesis fallthrough, non-divisor reject)
• Corpus parity: 168/197 excellent. lower-tf-probe-02-bool-array on 1m flips from NO_ENGINE_TRADES to excellent (3116/3116 trades, 0% diff). 15m baseline (lower-tf-probe-01) maintained at excellent 819/819.