feat(core): optimize character output to avoid deleting and reinserting identical content

Where a rule emits exactly the same characters as found in the context, it would be useful to avoid deleting and reinserting. For example the following rule should not require us to delete and reinsert `U+000A`:

```
U+000A U+094D > U+000A U+0905
```

The same applies for `context(1)`:

```
U+000A U+094D > context(1) U+0905
```

Apart from the marginal performance impact, this will increase compatibility, because we can handle control characters such as LF as shown above more seamlessly; see for example #14148 for an example of this problem.

As I currently remember it, LDML keyboards already do this, through the normalization module.