fix: honor lazy recursive EMA call clocks

[luisleo526]

Summary

• classify chart TA call sites across Pine v6 lazy and/or RHS and ternary branch edges
• route only recursive ta.ema sites on those lazy edges through their real call clock instead of eager precalculation
• preserve the accepted lazy-SMA scope, eager operand positions, unrelated TA families, request.security evaluator state, and distinct UDF call-site state

Validation

• focused lazy/precalc fences: 15/15
• full codegen suite: 1,554 passed, 1 expected skip
• explicit compile corpus: 254/254; base versus candidate corpus generated output changes: 0
• private impact inventory: 418 strategies, unchanged 411 OK / 7 errors, exactly 9 generated-output changes
• affected 9-row no-cache gate: 0 DOWN; all eight non-Quant engine tapes byte-identical
• Quant composition with merged Engine PR #109: strict 123 to 129/135, engine trades 246 to 200, countAbs 111 to 64, entry/exit/PnL P90 remain 0/0/0
• full forced-no-cache gate: 412/412 accounted, 404 OK + 1 no-trades + 6 transpile + 1 run error, canonical board 341/52/4/5, 0 UP / 0 DOWN
• no TradingView export, Pine source rewrite, tolerance change, or tier override