Preserve grouping and typing in TA reset length lowering

[luisleo526]

Summary

• The stable-runtime-TA-length reset path (PR Support stable runtime TA lengths #40) re-materialized length-defining expressions via a lossy AST->infix serializer: no grouping parentheses, no numeric coercion. C++ precedence re-associated the expression and int/int division truncated; a derived EMA length collapsed through sqrt(negative)->NaN->max(1, int(round(NaN))) to 1, silently degrading every TA instance built from it (strategy traded zero times).
• Fix: parenthesize non-atomic operands in both serializers so the expanded expression re-parses to the same tree, then lower the reset expression through the same expression visitor used for scalar statements — grouping and Pine numeric typing correct by construction. Input-backed vars render as override-aware get_input_*() reads in reset context (resets may run before input member init). Legacy renderer kept as fallback only.

Validation

• Full suite: 1443 passed / 0 failed / 2 pre-existing skips (incl. test_compile_corpus.py 254).
• New regressions: grouping-across-division; float division of two int operands in derived lengths.
• Independent fresh-context R7 gate: PASS (diff inspection, full suite re-run, e2e on the exposing strategy — derived length now 21 as Pine computes, 996 trades vs prior 0; an unaffected strategy transpiles byte-identical to main).

🤖 Generated with Claude Code