Change Symbol::as_str to &self -> &str

[oli-obk]

#76460 experimented with this, but it was never finished. We should figure out whether we actually want this

Pros:

• &str is nicer to work with than SymbolStr, e.g. avoids the need for some & and &* occurrences.
• It would be nice to eliminate SymbolStr.
• It avoids the lie that SymbolStr strings have a static lifetime, replacing it with a smaller lie that the strings have a lifetime tied to the lifetime of the input Symbol.

Cons:

• That lifetime is still a lie, and it's possible to cause crashes with this.
• The lie is less obvious in a vanilla &str than it is within a SymbolStr.
• We can't mark &str as !Send and !Sync.
  • this may be why SymbolStr (or what it was called previously) was introduced in the first place, because Zoxc did it together with the work on compiler parallelization.
• I haven't yet eliminated SymbolStr entirely. The to_stable_hash_key() is the hardest case.

[Kobzol]

I supposed that this was fixed by #91957?