[FIRRTL] Better Annotation Mutation APIs

The existing `AnnotationSet` and `Annotation` APIs are immensely better than working with raw dictionaries. However, these are difficult to mutate. When you need to mutate these, you usually wind up iterating over the whole `AnnotationSet`, rewriting one field of one annotation, and then creating a new dictionary. This is both tedious and inefficient.

First, we should provide some higher-level APIs for working with `AnnotationSet`. Ideas include:

- `AnnotationSet::map`
- `AnnotationSet::filter`
- `AnnotationSet::filter_map`

More mutable APIs are also reasonable, but I'm not sure exactly what to call them. Behind the scenes these will efficiently rewrite the underlying data structures to avoid dumb stuff like copying everything.

This _may_ require updates to enable better mutation of `DictionaryAttr` in upstream MLIR.

Start by reviewing the existing AnnotationSet and Annotation APIs and how they represent their underlying data. Clarify the desired map, filter, and filter_map semantics, then investigate whether upstream MLIR DictionaryAttr mutation support is required; done means an agreed, efficient mutation API with its behavior defined.

Written by the indexing model from the issue text.