fix(as370): &SYSLIST(n,m) reaches inside a sublist operand (#94)

[mgrossmann]

Closes #94.

&SYSLIST(n) is the n'th positional operand; &SYSLIST(n,m) is the m'th element of that operand's sublist. The resolver evaluated the whole subscript text as a single expression — eval_seta("1,1") stops at the comma and yields 1 — so the second subscript was dropped and the reference returned the operand entire, parentheses included.

Used as a name, the generated statement carried (ALPHA,8,GAMMA) and the module died of IFO016, an over-long name field. That diagnostic was correct about what it saw and pointed nowhere near the cause: nothing in the source has an over-long name.

Measured, not assumed

capture.py on mvsdev, JOB02901, RC 0 — deck and listing both committed (tests/ref/syslist.obj, tests/listref/ifox-listing-syslist.txt). The fixture probes a three-element sublist and a plain operand on purpose, because the plain one decides whether per-level substitution needs a special case:

	(n,1)	(n,2)	(n,3)
operand 1 = (ALPHA,8,GAMMA)	ALPHA	8	GAMMA
operand 2 = BETA	BETA	null

It does not: sub_elem already returns a value with no leading ( for index 1 and nothing beyond, which is exactly what the guest does. The deck is byte-identical through ESD and TXT, the END card differing only in the translator IDR as everywhere else.

The first capture attempt was rejected — 22 statements flagged, IFO006 UNDEFINED VARIABLE SYMBOL. Assembler XF requires LCLA/LCLC for a macro's local SET symbols and as370 accepts them undeclared. A separate leniency, noted in the commit and the listref README, not fixed here.

The change

One subscript per level, applying sub_elem once each. The scan is also balanced rather than strchr(')'): a subscript may carry its own parentheses — &SYSLIST((&I+1),1) — and the first ) is then in the wrong place. syslist_nested covers that on its own.

The named-parameter form &P(m) was already right and stays in the fixture as a control; it shares sub_elem and only ever needs one level.

sub_elem's tail copy moves from strncpy to the file's own scopy. gcc 14.2 reads the pair as a truncating copy once the new caller makes the sizes visible at the inline site — out holds 96 bytes and v can be the whole &SYSLIST buffer. The truncation is intended and bounded as it always was; scopy says so and terminates.

Verification

deck vs. IFOX00	byte-identical through ESD + TXT
tests/corpus/check.sh	no module moved
826 ecosystem modules	unchanged
tests/run.sh	49 OK, 0 failures, ALL SAMPLES BYTE-IDENTICAL TO IFOX00
tests/listref/check.sh	all references still column-exact

Both new cases fail against main (expected RC 0, got 8 / a parenthesised subscript truncated the expression).

On a local tree of real IBM MVS assembler — not committed here — 6 more of a 277-module spread sample assemble clean, none lost. Measured against main including #93, so this is #94's own contribution.

Built warning-clean with real gcc 14.2 -Werror on mvsdev as well as clang; it caught the strncpy truncation above, which clang accepts.