Make ANTLR parser run simulations end-to-end

[tmolteno]

Summary

The ANTLR parser (antlr/nec_parse) already dispatched all 21 program cards into the necpp engine, and RP/XQ already called simulate(). Five bugs stopped those simulations from running correctly and visibly. This PR fixes them so the parser actually performs simulations and emits their results.

Changes

File	Change	Why
antlr/nec_visitor.h	Added nec->calc_prepare() in visitGeCard	Lifecycle: buffers simulate() indexes (current_vector) were sized after the walk, but RP/XQ fire simulate() during it. Mirrors the geometry-only visitor and production driver order.
antlr/nec_visitor.h	Fixed dispatch_labeled to skip N leading tokens of rest_ints	Field mapping: its comment claimed ANTLR removes labelled tokens from ctx->INT() — false; getTokens() returns all tokens including labels. The mode integer (EX/GN/LD/NT/RP type) was consumed twice, shifting tag/seg and corrupting every labelled card.
antlr/NECFull.g4	Made imaginary voltage optional for EX type 0/5 (fnum fnum → fnum fnum?)	Grammar: the type-0 voltage-source branch mandated both real and imaginary voltage, rejecting valid minimal inputs like EX 0 0 4 0 1. (trailing blanks default to 0 in NEC-2).
antlr/nec_test.cpp	s_output.set_file(stdout) before set_output	Output dropped: the sink was never connected, so set_output cached a null FILE* and all simulation output was silently discarded. Uses set_file (not set_stream) because nec_printf/integer/real_out short-circuit on null m_output_fp.
antlr/nec_test.cpp	Removed post-walk calc_prepare(); added context.write_results(std::cout)	calc_prepare now runs in visitGeCard; write_results emits the structured result tables (radiation patterns, impedances, near fields) accumulated by RP/XQ/NE.
antlr/Makefile	Added end-to-end simulation smoke test (FR/EX/RP/EN) asserting radiation-pattern + power-budget tables present	Previously test: only checked parsing/geometry, never a real simulation.
antlr/build_in_docker.sh	Generates ../build/simple/config.h (NECPP_VERSION/NECPP_BUILD_DATE)	So the antlr build does not depend on a prior top-level make.

No changes to src/ — the engine and the shared nec_card_parser.h dispatch table were already the correct integration surface.

Verification

• example1.nec (EX + XQ + NE) simulates; near-E field values match production nec2++ numerically (e.g. INPUT POWER = 4.4773E-03 Watts, near-E field 5.5442E+01 -66.31 at 0.0179 m).
• Full make test passes: geometry, all card types, and the new simulating case (6249 bytes output, both tables present).
• The full-card-set case still prints one CHECK DATA error — confirmed it is NT 0 1 0 0 … with a placeholder second-port segment of 0, which production nec2++ rejects identically (invalid input data, not a parser defect). That test tolerates it via || true.

Out of scope

CLI flags (-i/-o/-s), NX multi-job looping, PL filename handling, full nec2++ drop-in parity.