fix: guard link recognizer against stale/detached node in _tapNodeLink

[mtallenca]

Description

_TextLineState caches per-node link gesture recognizers in _linkRecognizers (Map<Node, GestureRecognizer>), and the recognizer's callback captures a specific segment node:

_linkRecognizers[segment] = TapGestureRecognizer()
  ..onTap = () => _tapNodeLink(segment);

That map is only cleared on a readOnly toggle (didUpdateWidget), a meta/ctrl key change (_pressedKeysChanged), or dispose — never when the line's document content changes. So the cached recognizer (and the node it captured) can outlive a document mutation/re-render.

When the gesture finally fires, _tapNodeLink/_longPressLink re-read the node's link attribute and force-unwrap it:

void _tapNodeLink(Node node) {
  final link = node.style.attributes[Attribute.link.key]!.value; // throws if null
  _tapLink(link);
}

If the document mutated between recognizer creation and the tap being swept by the gesture arena, the captured node can be detached or no longer carry a link attribute, so attributes[Attribute.link.key] is null and the ! throws.

We see this crash in production (Crashlytics, flutter_quill 11.5.1, Android) on a read-only document with many links that is re-rendered while the user navigates — a queued tap is delivered during GestureArenaManager.sweep against a now-stale recognizer:

Fatal Exception: Null check operator used on a null value
0  _TextLineState._tapNodeLink (text_line.dart:670)
1  _TextLineState._getRecognizer.<fn> (text_line.dart:656)
2  GestureRecognizer.invokeCallback (recognizer.dart:345)
3  TapGestureRecognizer.handleTapUp (tap.dart:758)
4  BaseTapGestureRecognizer._checkUp (tap.dart:383)
5  BaseTapGestureRecognizer.acceptGesture (tap.dart:353)
6  GestureArenaManager.sweep (arena.dart:173)

Fix

Read the link attribute defensively (?.value) in both _tapNodeLink and _longPressLink instead of force-unwrapping. _tapLink already no-ops on a null link; _longPressLink gets an early return. When the captured node is stale, the gesture simply does nothing instead of crashing.

Notes

This is a defensive guard against a timing/staleness window (stale cached recognizer fired during arena sweep after a document mutation), which is awkward to reproduce deterministically in a widget test. Happy to add a regression test if a maintainer can point me at the preferred way to drive a cached recognizer against a mutated node.