fix(parser): support simple/series/const UDF param qualifiers and T[] postfix-array decls

[luisleo526]

Problem

Two parser defects surface when transpiling real-world published PineScript v6 strategies (found while backtesting scraped TradingView strategies through the engine):

1. UDF parameter type qualifiers dropped. Only series was special-cased in _parse_func_def, so simple/const fell through. ma(float s, int l, simple string m) => parsed as 4 params (s, l, simple, m), and the generated C++ failed to compile:

   error: no matching function for call to 'ma_cs0(double&, int&, std::string&)'
   note: candidate: 'double ma_cs0(double, int, std::string, double)'
   

2. Postfix-array declarations silently dropped. var float[] qp = array.from(...) / float[] x = ... produced no VarDecl at all — _parse_type_hint_string left the trailing [] unconsumed, the name failed to parse, and statement-level error recovery discarded the whole line. Downstream every reference became Unknown variable 'qp'. (Untyped var qp = array.from(...) worked, masking it.)

Fix

• Consume any leading series / simple / const qualifier before the (optional) type in UDF params. Also fixes UDT-typed series/simple params.
• Parse the T[] postfix as array<T> in _parse_type_hint_string, and recognize the bare T[] name = form in the statement dispatcher.

Tests

5 regression tests in tests/test_parser.py (param qualifiers single-param; postfix-array var/bare register a VarDecl; end-to-end transpile). Full suite green (pre-existing test_matrix_*_compiles env failures unaffected).

Regression verification (transpiler + engine)

• Isolation diff: re-transpiled all 258 corpus probes with this change vs. with it stashed → byte-identical generated.cpp for every probe (0 drift).
• Full local build + run: built runtime + 246 strategy .so, ran the sweep → ok=246 fail=0, verify_corpus --all = 245 excellent / 1 anomaly (unchanged canonical baseline); regenerated engine_trades.csv byte-identical to committed.

🤖 Generated with Claude Code