c_geometry: write parsed fields back on trailing-field cards

[KJ7LNW]

Description

The shared geometry card parser only wrote its output pointers inside the in-loop end-of-string guards. A card carrying more numeric fields than the two integers and seven reals the reader consumes left line_idx short of the terminator, so the write-back never fired and the caller (which reuses one parse state across every card) retained the previous card values.

Concretely a GM move card took its repeat count from a preceding GA arc segment count and replicated the whole structure, exploding the segment total. The regression came from 220977d, which extracted the shared parser and dropped the unconditional write-back that the nec2c and xnec2c ancestors retain.

Type of Change

• Bug fix (non-breaking change that fixes an issue)

Implementation Details

Restore the unconditional write-back of the parsed integers and reals at the normal loop-exit of the shared geometry card parser, repairing every geometry card type that carries trailing fields.

Testing

Reproduce with cebik 84-8.nec, whose 12-field GM card after two 90-segment GA arcs yielded 16380 segments instead of 180 and ran multi-hour at 8 GB:

CM 2-element circular quad beam
CE
GA 1 90 .1572 0 360 .001
GA 2 90 .1722 0 360 .001
GM 0 0 0 0 0 0 .1665 0 2 0 0 0
GE 0 -1 0
EX 0 1 68 0 1.0 0.0
FR 0 1 0 0 299.8 1
RP 0 1 361 1000 90 0 1 1
EN

[tmolteno]

Review ✅ Approved & Merged

@KJ7LNW — thank you for this excellent contribution. This is a textbook
bug fix: single logical correction, clear root cause, concrete reproducer
(84-8.nec), zero risk to existing paths, and a severe regression
(16K→180 segments on a GM card).

Key findings from review:

• Root cause precisely identified: the shared parse_geometry_card_line()
  only wrote back on \0 terminator, but GM has 12 fields (exceeding the
  2+7 parsed), so the write-back never fired and the caller's reused parse
  state retained stale GM repeat-count values from the preceding GA card.
• Fix is idempotent: the unconditional write-back at function-end fires
  only on normal loop exit. All existing working paths return before reaching
  it, so zero risk to existing callers.
• Excellent reproduction deck: the Cebik 84-8.nec quad beam clearly
  demonstrates the segment explosion (180→16380 before, 180 correct after).

Nice work!