🪤 Audit semver traps in viceroy-lib

[acfoltzer]

While reviewing the release branch in #261, I noticed that we have an awful lot of viceroy-lib exposed as public, including session methods like insert_pending_request() and types like AsyncItem. It is very difficult to imagine these being useful to anyone outside of viceroy-lib itself, and so their presence in the public interface is a semver trap that I don't think yields us much benefit.

In addition, we have types like Error which are legitimately useful by library clients, but that are missing defensive measures like #[non_exhaustive] to prevent breakage via adding variants.

I think we should either:

• reduce our exposure to semver traps by
  • making the public interface of viceroy-lib smaller, and
  • using mitigations like #[non_exhaustive] to make the remaining public interface more forwards-compatible
• or adopt a wasmtime-like strategy where every non-bugfix release is a major semver bump, and we just don't worry about what we expose

I don't have a great sense of how many clients of viceroy-lib there are other than viceroy-cli, so maybe folks could weigh in here if either of these changes would impact you or others you're aware of?