[fly1d]

FLAG_IGNORE_SLICE_ORDER can return more than MaxDiff differences: with
the default limit of 10, two disjoint eight-element slices return 16.

Check the positive budget in the value-count loop so both passes share the
limit, including differences already found in earlier fields. Zero and
negative limits retain their existing unordered behavior. The original
tests are unchanged; 11 appended cases cover the limit and its controls.
The pending Differ work in #71 also retains the uncapped counting loop.

Validation on Go 1.26.6, macOS arm64:

• The same 11 cases on unchanged upstream: 6 cap failures, 5 controls pass;
  all 11 pass with the fix.
• go build, full tests with coverage, go vet ./..., and
  go test -race ./... pass. Core package coverage is 100.0%.
• Tracked Go formatting and git diff --check pass; modules and CI files
  are unchanged.

The existing Go test workflow filters master while this targets main;
these are local results, not a claim that the hosted Go/platform matrix ran.

[daniel-nichter]

This raises another issue: MaxDiff <= 0 actually produces zero diffs. That's the original because behaviour because, unfortunately, I never specified what happens if it's <= 0. So line 520 should be

if len(c.diff) >= MaxDiff {

like original code, and this test changed to wantCount: 0. That would truly preserve original, undefined behaviour.

Thoughts?