do178: text-mode summary line prints negative GAP count (-1 GAP) on DAL-B report #86

What's wrong

`gofusa do178` (v0.47.0, and reproduced back to at least v0.36.0) prints a nonsensical negative GAP count in its text-mode summary line for a DAL-B report:

```
DO-178C Gap Report
Project: go-LIN DAL: DAL-B Generated: 2026-07-29

Summary: 39 objectives 30 PASS -1 GAP 9 MANUAL 1 N/A 0 FAIL
```

Counting the actually-printed per-objective lines in that same report: 30 PASS, 7 MANUAL (not 9), 1 N/A, 0 FAIL — 30+7+1 = 38, leaving 1 real GAP, not -1. The header's own arithmetic (39 total - 30 PASS - 9 MANUAL - 1 N/A - 0 FAIL = -1) doesn't match what's actually enumerated below it; the "9 MANUAL" figure in the header appears to be counting something extra that isn't reflected in the individual ? [...] MANUAL lines.

Evidence it's the text renderer, not the underlying data

The same report via -format json has an internally consistent summary block with no negative values:

```
$ gofusa do178 -format json | jq .summary
{
"total": 37,
"satisfied": 23,
"partial": 7,
"gaps": 7
}
```

(Note even the total differs between text mode, 39, and json mode, 37 — the two renderers appear to be counting objectives differently, not just formatting the same numbers differently.)

Repro

Against github.com/SoundMatt/go-LIN main:

```
go install github.com/SoundMatt/go-FuSa/cmd/gofusa@v0.47.0
git clone https://github.com/SoundMatt/go-LIN
cd go-LIN
gofusa do178 # text summary shows "-1 GAP"
gofusa do178 -format json | jq .summary # internally consistent, no negatives
```

Why it matters

A negative count of unmet objectives in a DO-178C compliance report is nonsensical on its face and undermines confidence in the DO-178C evidence bundle specifically for anyone reading the text-mode report (which is what a human auditor is most likely to look at). It isn't currently caught by any go-FuSa self-check, and gofusa do178's own exit code doesn't reflect it either (exits 0 regardless).

Suggested fix

• Reconcile the text-mode summary-line counters with the same counting logic already used for the -format json summary (or however the JSON summary's satisfied/partial/gaps/total map onto the text mode's PASS/GAP/MANUAL/N/A/FAIL categories) so both renderers agree, and neither can go negative.
• Consider a self-test that asserts gofusa do178 -format text|json on the go-FuSa module's own project (dogfooding) produces non-negative summary counts, so this class of regression is caught in go-FuSa's own CI.

Found while doing a scheduled go-FuSa version-currency bump on github.com/SoundMatt/go-LIN (v0.36.0 → v0.47.0); see SoundMatt/go-LIN#53.