Rewrite `getAccessibleSymbolChain` for performance

getAccessibleSymbolChain is one of the oldest parts of the compiler - remaining today mostly unchanged (barring support for new features) from when it was used in the old text-based declaration emitter. getAccessibleSymbolChain is used to, given a [symbol, scope] pair, find a series of symbols whose exports can be accessed to lookup the symbol. Today, this is by and large uncached, so the fail case, when a symbol is not accessible, causes a traversal of every publicly reachable symbol in a program - this process is then repeated for every symbol that needs to be named, which means you end up spending a very long time traversing symbol structures. On normal typescript code, the inefficiency is less outsized, as we only need to invoke this code when we generate inferred types in declaration emit, which is more rare in TS output than in JS output, however you can still find projects where its' influence is large (some of @AnyhowStep 's samples bad performance in declaration emit trace back to this).

As far as the direction of the fix goes - I have some ideas. Today, we enumerate all possibilities and just backtrack to see what works - instead, I imagine using a (cacheable) set-like structure to check if the symbol is even accessible from a given symbol, and then use a hierarchy of these sets to guide chain creation.

Commencez par localiser getAccessibleSymbolChain dans le compilateur et par suivre ses appelants pendant le declaration emit. Examinez le parcours de symboles existant ainsi que les traces de performance mentionnées dans l’issue, puis déterminez comment valider que les parcours répétés sont réduits sans modifier les résultats de la accessible symbol chain.