Fix TwoWayDict reverse map going stale on key reassignment

[c-tonneslan]

TwoWayDict.__setitem__ writes the new value into _forward and _reverse, but never removes the old value from _reverse when a key is reassigned:

d = TwoWayDict({})
d[1] = 10
d[1] = 20
d.get(1)               # 20  ✓
d.get_key(20)          # 1   ✓
d.get_key(10)          # 1   ✗ should be None
d.contains_value(10)   # True ✗ should be False

So _reverse stops being a true inverse of _forward — get_key and contains_value keep reporting a value the key no longer maps to.

The fix drops the old value's reverse entry when the key already exists. The pre-existing TODO about duplicate values (two keys → one value) is a separate design question and is left untouched.

Added test_set_item_reassigning_key_clears_old_reverse_entry alongside the existing test_set_item.

Worth noting on scope: DataTable (the in-tree user of TwoWayDict) currently rebuilds these dicts rather than reassigning keys, so this isn't a user-visible bug today — it's a correctness hole in the utility class itself, which has an explicit two-way invariant and its own test file.

[willmcgugan]

Your PR has been closed due to a AI policy violation.

Please read the following before submitting further PRs.

https://github.com/Textualize/textual/blob/main/AI_POLICY.md