Provide a way to define labels with regular expression replacement patterns

Right now, we are able to define simple lexical patterns like this:

"% biome" to define classes like "desert biome". Now the interesting thing is that in the case of biomes we define them in terms of "ecosystems", which means that the filler class is called something like "desert ecosystem". Now, we don't want to label our new class "desert ecosystem biome" - that's silly.

What would be good if we had a more flexible way to specify patterns for labels and definitions. For example something like:

```
text: "% biome"
match: r'(.*) ecosystem (biome)'
replacement: "$1 $2"
```

This would not be used for inference, just as a postprocessing step after the label is generated. I guess one may want to provide multiple match and replace rules, so his would be more flexible:

```
text: "% biome"
postprocessing:
- match: r'(.*) ecosystem (biome)'
replacement: "$1 $2"
```

@cmungall sounds a bit too crazy?

The issue names no implementation files, tests, or entry points. Start by locating the existing lexical label-pattern handling and its configuration schema, then determine how postprocessing rules would be represented and applied; done means supporting the shown match and replacement behavior without using it for inference.

Written by the indexing model from the issue text.