…(helper inlining)

Two coupled fixes let a request.security nested in a multi-call helper transpile
+ compute correctly. Both are additive with a byte-identical fallback, so every
already-transpiling strategy is unchanged (corpus 265/265 byte-identical).

1. History index (security.py). `f_secure_ema(tf, length, idxHigher, idxCurrent)
   => request.security(sym, tf, ta.ema(close, length)[idxHigher])[idxCurrent]`
   rejected because the TA history index `[idxHigher]` is a helper param, not a
   NumberLiteral. Add _resolve_security_index_literal: try the existing literal
   check first (short-circuit → identical), else resolve the index through the
   helper binding stack / global_expr_map and constant-fold batch-constant
   barstate flags (_fold_security_const_bool: isrealtime/islast→false,
   ishistory→true) — matching the flag folding codegen already does elsewhere.

2. Constructor length (base.py). A request.security in a helper called N times is
   cloned per call site (distinct sec_id + callsite_idx), but the shared TA
   site's ctor_args were resolved ONCE against call site 0 — so four security
   EMAs all sized from crossFastLen (21) instead of the per-site [21,55,21,55].
   Route each sec's ctor-arg resolution through the existing per-call-site
   function-clone TA remap (_func_cs_ta_remap); identity for cs0/non-clones.

Recovers officialjackofalltrades-concordance-execution-mandate-joat from
transpile-error → STRONG (was: could not transpile). Entry/exit price p90 both
0.0000% exact, countAbsDelta 2 (residual is strategy-logic trade-count precision,
not codegen — excellent needs exact count parity). pytest 1446 passed; clang
exit 0.

Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>