[NFCI] Refactor FormatterContainerPair into TieredFormatterContainer.

`FormatterContainerPair` is (as its name indicates) a very thin wrapper
over two formatter containers, one for exact matches and another one for
regex matches. The logic to decide which subcontainer to access is
replicated everywhere `FormatterContainerPair`s are used.

So, for example, when we look for a formatter there's some adhoc code
that does a lookup in the exact match formatter container, and if it
fails it does a lookup in the regex match formatter container. The same
logic is then copied and pasted for summaries, filters, and synthetic
child providers.

This change introduces a new `TieredFormatterContainer` that has two
main characteristics:

- It generalizes `FormatterContainerPair` from 2 to any number of
  subcontainers, that are looked up in priority order.
- It centralizes all the logic to choose which subcontainer to use for
  lookups, add/delete, and indexing.

This allows us to have a single copy of the same logic, templatized for
each kind of formatter. It also simplifies the upcoming addition of a
new tier of callback-based matches. See
https://discourse.llvm.org/t/rfc-python-callback-for-data-formatters-type-matching/64204
for more details about this.

The rest of the change is mostly replacing copy-pasted code with calls
to methods of the relevant `TieredFormatterContainer`, and adding some
methods to the `TypeCategoryImpl` class so we can remove some of this
copy-pasted code from `SBTypeCategory`.

Differential Revision: https://reviews.llvm.org/D133910