PrefixContainer.GetKeysFromPrefix does a full linear scan instead of using the existing sorted array

`PrefixContainer` already builds a sorted copy of all form keys in the constructor (used by `ContainsPrefix` via binary search), but `GetKeysFromPrefix` ignores it and iterates `_originalValues` linearly every time it's called. In case of nested collection models with inner `Dictionary<>` call is repeated, so should be optimized.

The fix would be to use `_sortedValues` in `GetKeysFromPrefix` — binary search to find the start of the prefix range, then scan forward only while entries still match, reducing each call from O(N) to O(log N + M) where M is the number of matching entries.

Start at PrefixContainer.GetKeysFromPrefix and compare its current iteration with ContainsPrefix, which already uses the sorted values and binary search. Trace _sortedValues and _originalValues to confirm the prefix-range behavior; done means matching keys are returned with existing behavior while the lookup uses the sorted range rather than a full scan.

Written by the indexing model from the issue text.