fix: omit partial initial daily bars from OTC security requests

[luisleo526]

A forex or CFD intraday run starting mid-session counted an incomplete daily bar toward request.security indicator warmup. Daily ATR could therefore become available one session too early. The engine now excludes that initial partial daily bucket on the single-feed historical path and applies the same boundary to lookahead projections, using the actual session open to preserve OANDA's earlier daily timestamp.

Validation on engine e729b1432c1a5fdc7e95ac33d1e879232b55e37f with codegen 3fd97fe28abd7b191cb376d9f5db4e056fcfae4b:

• Cloud Run measured all 4,190 probes in 72/72 cases. Asian Box on XAUUSD moves from moderate to excellent: canonical match 94.4% to 100%, count difference 1 to 0, and exit/PnL P90 errors to 0. The other 4,189 canonical results are unchanged, with zero regressions, errors or gaps. Result: 4,160 excellent / 30 strong / zero moderate.
• Formal gate pineforge-pr-gate-4686q: PASS, target score +2, and zero regressions or tolerated exceptions across 704 hard probes.
• Fresh build and 212/212 tests passed. Covered TradingView controls and synthetic fixtures verify the first daily bucket, ATR14 history, lookahead visibility, and forex/CFD session boundaries. Independent Grok 4.6 review of this exact candidate is GREEN, P0/P1/P2 = 0; all prior findings are resolved.

The target has 244 trades and 488 entry/exit rows; 480 economic rows match TradingView exactly and eight residual timing/price differences remain. The excellent grade does not imply complete economic identity.

Grading metrics, thresholds, tolerances, verifier selection, harness, tapes and population are unchanged. Existing native/auxiliary feed, weekly/monthly, explicit warmup and streaming rules remain in place.

Candidate snapshot: 73babb95ef95308dff77370d9babb008ab0818217c4513363d276ad54d5d8796 (788,852 bytes).