Refactor Encoding to split fast-path and fallback logic (#23098)

Refactor Encoding to split fast-path and fallback logic (#23098)

This refactoring is limited to ASCIIEncoding at the moment, but it can easily be applied to UTF-8 / UTF-16 / UTF-32.

High-level changes:
- Fallback logic has been split from the fast-path, improving performance of GetBytes and similar routines.
- All of the plumbing of when to invoke the fallback logic and how to manage leftover data has been moved into the base class.
- Almost all of the logic except for the fast-path is now written in terms of verifiable code (Span and ReadOnlySpan).
- Minor bug fixes in EncoderNLS.Convert (see https://github.com/dotnet/coreclr/issues/23020).