fix: prevent NaN fling direction when tracked pointer offsets are zero

[nisenbeck]

Fixes #2221 - see that issue for the full bug report,
reproduction steps and root-cause analysis.

Fix

MapInteractiveViewerState._handleScaleEnd could divide flingOffset by a
zero flingDistance (0.0 / 0.0 = NaN) when both the final tracked pointer
segment and the overall drag offset happened to have zero length, even
though the gesture's reported velocity was above the fling threshold.

Extracted the direction calculation into a small, pure, unit-testable
flingDirection helper, and added a final fallback to the (already
guaranteed non-zero) velocity direction instead of dividing by zero:

if (finalSegmentDistance > 0) return finalSegment / finalSegmentDistance;
if (flingDistance > 0) return flingOffset / flingDistance;
return velocityDirection;

This preserves the existing behaviour (prefer the final tracked segment,
then the overall drag direction) and only changes behaviour for the
previously-undefined zero/zero edge case, where it now falls back to the
same velocity-based direction that flutter_map used prior to #2158 (8.2.2
and earlier).

velocityDirection is guaranteed finite: magnitude is already checked to
be >= _kMinFlingVelocity before this is called, and Flutter's own
VelocityTracker / LeastSquaresSolver never return NaN - they fall back
to Offset.zero / null on ill-conditioned samples - so this can never
itself be a source of NaN.

Testing

• Added test/gestures/map_interactive_viewer_test.dart covering all three
  branches of flingDirection, including a regression test for the
  zero/zero case.
• Full existing test suite passes locally with no regressions.
• Manually verified on a physical Android device: confirmed the crash on
  8.3.0/8.3.1, confirmed it does not occur on 8.2.2, and confirmed this
  fix resolves it while pointing a reproduction app at this branch via a
  dependency_overrides: path: entry.

AI disclosure

Per CONTRIBUTING.md: I used AI assistance (Claude Code) to help diagnose
the root cause and draft this fix and its tests. I've reviewed the change,
verified it against the project's lint rules and test suite, and manually
confirmed the fix resolves the crash on a physical device.

[JaffaKetchup]

Hey, thanks for this. Please open a bug report issue, and move your details of the issue there - this helps with tracking. The PR should only detail the fix. It would also be good to get a screen recording of you reproducing this issue.

[nisenbeck]

Thanks for the quick feedback! I've split this up as requested:

• Bug report with full reproduction steps, root-cause analysis and a screen recording: Fling gesture with very close extreme points causes division by zero #2221
• This PR description now only covers the fix itself

Let me know if anything else is needed.

[JaffaKetchup]

LGTM!

Really not sure how you ran into this bug in practise since it was quite difficult to reproduce, but I guess it is the sort of thing that seems to just randomly happen. I also would expect the offset never to be zero since I would've expected that to require more precision than is available, but I guess it happens.

From what I can tell, this fixes the bug. Thanks :)

[nisenbeck]

Thanks for merging! Found it while checking whether #2199 ("LatLng is not finite") was actually resolved on 8.3.1 - was just doing normal pan/fling testing on the map and hit this and a separate issue (#2222), unrelated to #2199 itself.