Order scopes on save, and optimize the happy path for access token read - #644

This PR currently contains 2 commits.

1. The first commit orders scopes on save, which is similar to Order scopes on save to avoid access token duplication in the cache microsoft-authentication-library-for-dotnet#4479. Existing unit tests need adjustment to match the changed scope order.
2. The second commit optimizes the happy path of access token read, which was the idea mentioned in this conversation. Basically, when searching for an access token, this PR uses the token key to attempt an O(1) search, and falls back to O(n) search when necessary.
   • No other unit test adjustment is needed, meaning this change is backward compatible.
   • Existing benchmark test shows ~40% time reduction (or 1.6x speed improvement) on "token cache hit" code path, which is also visible in diagram No. 2 and No. 4 in this benchmark page. (Note that the performance gain can be much higher than 1.6x if each tenant has more tokens; it is actually an O(1) vs O(n) improvement.)